CN102769753B - H264 encoder and coding method - Google Patents

Abstract

The present invention provides an H.264 encoder and encoding method. When realizing inter-frame prediction, by setting an appropriate search window and decoding the reference code stream, the reference data required by the current macroblock is obtained. The involved The storage space only includes code streams for reference and part of the reference data, and the size is much smaller than the size of a frame of image, thus effectively reducing the storage space of the reference frame required by the H.264 encoder; at the same time, because no off-chip memory is used, The cost of the entire H.264 encoder and system power consumption are effectively reduced. Moreover, since the read and write performance of on-chip storage is generally much better than that of off-chip storage, when encoding large-size images, image quality can be guaranteed. At the same time, the storage data throughput rate is greatly improved.

Description

H264 encoder and encoding method

technical field

The present invention relates to the technical field of image processing, in particular to an H.264 encoder and an encoding method.

Background technique

The H.264 standard is a high-performance digital video codec standard proposed by the Joint Video Team (JVT, Joint Video Team). Its biggest advantage is its high data compression rate. Under the condition of the same image quality, the compression ratio of H.264 can reach twice that of MPEG-4. The reason for such a high compression ratio is that H.264 contains a series of new features, such as multi-frame reference , variable block size motion compensation, and high-precision sub-pixel motion compensation, deblocking filter, etc.

According to the H.264 standard, macroblock (Macroblock or MB for short) types are mainly divided into two categories: intra-frame (Intra) coding and inter-frame (Inter) coding. Intra-frame encoding uses the adjacent data of the current macroblock for prediction, and then subtracts the predicted value from the current macroblock data to generate residual data, and finally performs subsequent encoding processing on the residual, without using other images in the whole process The data. However, inter-frame coding needs to use the encoded reconstructed image to make predictions, that is, to search for the data block that best matches the current block in the reconstructed image. Due to the temporal redundancy of video images, inter-frame prediction can often find a better Intra-prediction better matches the data of the current block, thus improving the overall image compression performance. It can be seen that the use of inter-frame coding can improve the compression performance of video images, but a certain storage space is required to store some reconstructed images, or referred to as reference frame (reference frame) images, when performing inter-frame prediction. According to the H.264 standard, the reference frame is at least 1 frame and can be up to 16 frames. If the size of the image to be encoded is large (such as VGA: 640x480), then the storage space for this reference frame is still considerable.

As shown in Figure 1, the currently disclosed H.264 encoder includes the following submodules:

CurrentFrame: Input the current image data;

ME/MC: MotionEstimation/MotionCompensation, which realizes inter-frame prediction;

ChooseIntraMode: Choose the best intra prediction mode by comparing the encoding cost;

IntraPrediction: Realize intra-frame prediction according to the best intra-frame mode;

ModeDecision: Choose whether to use intra prediction or inter prediction according to the size of the encoding cost;

DCT/Q: Realize forward DCT transform (that is, integer discrete cosine transform) and quantization;

EntropyEncoder: Entropy encoding module;

IDCT/InvQ: Realize inverse DCT transformation and inverse quantization;

Deblocking: Filter the image in order to reduce the block effect;

Off-chip memory: Provide reference frame data 1~16ReferenceFrames of 1~16 frames.

The H.264 standard uses integer discrete cosine transform (ie, DCT transform) and quantization to eliminate the correlation in the image signal and reduce the dynamic range of image coding to achieve the purpose of compression. In Fig. 1, after the residual data Dn obtained from the current frame undergoes forward DCT transformation and quantization processing, on the one hand, it enters the entropy coding module for coding, and then gives the code stream (bitstream) after image compression; on the other hand, , after entering the inverse DCT transformation and dequantization, plus the predicted value, the reconstructed data Fn' is obtained. The reconstructed data can be used as the adjacent block value to participate in the intra-frame prediction of the current frame, and they are filtered by the loop Deblocking It can be written into the off-chip memory as reference data for inter-frame prediction of subsequent frames.

Although the H.264 encoder in the prior art can realize inter-frame prediction by using an off-chip memory to meet the high requirements for memory space and data throughput when encoding large-size images, on the other hand it will High data throughput greatly increases system cost and system power consumption, making the H.264 encoder unable to meet the requirements of low-power devices.

Contents of the invention

The purpose of the present invention is to provide an H.264 encoder and encoding method, which can save the storage space of reference frames, improve data throughput, and reduce system cost and power consumption while ensuring image quality.

In order to solve the above problems, the present invention proposes a H.264 encoder, including:

The encoding subsystem is used to input the data of the current image, perform intra-frame prediction or inter-frame prediction of the current macroblock to obtain the residual data of the current macroblock, and perform DCT transformation, quantization and entropy coding on the residual data Then output the code stream of the current macroblock, wherein, when performing inter-frame prediction on the current macroblock, set the search window of the current macroblock, and prepare the required reference data in the search window;

A code stream caching module, configured to cache a reference code stream, where the reference code stream is an encoded code stream of a reference frame image;

The decoding subsystem is used to perform entropy decoding, reverse DCT transformation and inverse quantization on the reference code stream, and obtain the reference data required by the encoding subsystem for inter-frame prediction;

The reference data buffering module is used for buffering the reference data obtained by the decoding subsystem, and providing the encoding subsystem with the reference data required in the search window during the inter-frame prediction of the current macroblock.

Further, the encoding subsystem includes:

The control unit is used to input the current macroblock data, and send a macroblock start signal to other units to notify other units to start the corresponding calculation of the macroblock;

The residual unit is used to perform intra-frame prediction or inter-frame prediction of the current macroblock to obtain the residual data of the current macroblock, wherein, when performing inter-frame prediction on the current macroblock, set the current macroblock Search the window, and prepare the required reference data in the search window, the reference data is obtained by decoding the reference code stream;

A forward DCT transform and quantization unit, configured to perform forward DCT transform and quantization on the residual data;

An entropy encoding unit, configured to perform entropy encoding on the forward DCT transformed and quantized residual data to generate a code stream of the current macroblock;

Inverse DCT transformation and dequantization unit, for performing reverse DCT transformation and dequantization on residual data after forward DCT transformation and quantization;

The reconstruction unit is used to add the residual data after inverse DCT transformation and inverse quantization to the intra-predicted or inter-predicted data of the current macroblock to obtain reconstructed data, which is the next macroblock The intra prediction of provides reference data.

Further, the residual unit includes:

A prediction mode selection module, configured to select an intra-frame prediction or an inter-frame prediction mode for the current macroblock according to the size of the encoding cost;

The inter-frame prediction module is used to set the search window for the current macroblock in the inter-frame prediction mode, select the reference data required by the search window from the reference data cache module, and pair the search window according to the reference data The current macroblock performs inter-frame prediction;

An intra-frame prediction mode selection module is used to select the best intra-frame prediction mode of the current macroblock according to the reconstructed data obtained by the previously coded macroblock;

An intra-frame prediction module, configured to perform intra-frame prediction on the current macroblock according to the optimal intra-frame prediction mode and the reconstructed data;

The multiplex selection module is used to realize the selection of the prediction data output by the prediction mode selection module, the intra-frame prediction module and the inter-frame prediction module.

Further, the decoding subsystem includes:

an entropy decoding unit, configured to perform entropy decoding on the reference code stream;

An inverse DCT transform and inverse quantization unit, configured to perform inverse DCT transform and inverse quantization on the residual coefficients output by the entropy decoding unit;

an intra prediction unit, configured to perform intra prediction according to the intra prediction mode output by the entropy decoding unit;

The reconstruction unit is used to add the residual data output by the inverse DCT transform and inverse quantization unit of the decoding subsystem and the prediction data output by the intra prediction unit to obtain reconstructed data, which is used as the macroblock inter-frame prediction search The reference data required in the window.

Further, the intra-frame prediction unit of the decoding subsystem and the intra-frame prediction module of the encoding subsystem are the same intra-frame predictor or two independent intra-frame predictors that are multiplexed.

Further, the inverse DCT transform and inverse quantization unit of the decoding subsystem and the inverse DCT transform and inverse quantization unit of the encoding subsystem are the same inverse DCT transform/dequantizer or two independent The inverse DCT transform/inverse quantizer.

Further, the decoding subsystem further includes: a filtering unit, configured to filter the reference data obtained by the decoding subsystem before being cached in the reference data caching module.

Further, the reference data cache module is a circular memory, which is provided with a storage space of a certain number of macroblocks. When buffering the reference data obtained by the decoding subsystem, the currently obtained reference data covers the circular memory for inter-frame Reference data that will no longer be used for forecasting.

Further, the calculation formula of the number S of macroblocks stored in the reference data cache module is:

S=Ceil(sw_height_by_pix/16)*pic_width_by_mb-(pic_width_by_mb-Ceil(sw_width_by_pix/16))+1

Among them, Ceil(x) represents the smallest integer greater than or equal to x;

sw_height_by_pix is the height of the search window expressed in pixels;

sw_width_by_pix is the width of the search window expressed in pixels;

pic_width_by_mb is the width of the picture expressed in units of macroblocks.

Further, the reference data cached by the reference data caching module also includes P frame data, and the P frame data is obtained by encoding an I frame as reference data through the encoding subsystem and decoding through the decoding subsystem of.

Further, when the encoding subsystem encodes the subsequent P frames of the input current image, the I frame data or the P frame data in the reference data buffer module is used as reference data.

Further, the decoding subsystem includes a P frame decoding system for providing reference data of the P frame type and an I frame decoding system for providing reference data of the I frame type.

Further, the I frame decoding system includes:

A first entropy decoding unit, configured to perform entropy decoding on an I-frame reference code stream in the reference code stream;

A first inverse DCT transform and inverse quantization unit, configured to perform inverse DCT transform and inverse quantization on the residual coefficient output by the first entropy decoding unit;

a first intra-frame prediction unit, configured to perform intra-frame prediction according to the intra-frame prediction mode output by the first entropy decoding unit;

The first reconstruction unit is configured to add the residual data output by the first inverse DCT transform and inverse quantization unit and the prediction data output by the first intra prediction unit to obtain the I frame reference code stream The reconstructed data is used for the decoding of the P-frame reference code stream or as the reference data required in the search window for the inter-frame prediction of the current macroblock;

The first filtering unit is configured to filter the reference data of the first reconstruction unit.

Further, the P frame decoding system includes:

A second entropy decoding unit, configured to perform entropy decoding on the P frame reference code stream in the reference code stream;

A second inverse DCT transform and inverse quantization unit, configured to perform inverse DCT transform and inverse quantization on the residual coefficients of the second entropy decoding unit;

a second intra-frame prediction unit, configured to perform intra-frame prediction according to the intra-frame prediction mode output by the second entropy decoding unit;

a second inter-frame prediction unit, configured to perform inter-frame prediction according to the reconstructed data of the first reconstruction unit and the motion vector output by the second entropy decoding unit;

The second reconstruction unit is configured to add the residual data output by the second inverse DCT transform and inverse quantization unit to the prediction data output by the second intra prediction unit or the second inter prediction unit, to obtain the The reconstructed data of the P-frame reference code stream is used for the reference data required in the search window of the current macroblock during inter-frame prediction;

The second filtering unit is configured to filter the reference data of the second reconstruction unit.

Correspondingly, the present invention also provides a kind of H.264 coding method, comprises the following steps:

(1) Determine whether the current macroblock needs to perform inter-frame prediction, if no inter-frame prediction is required, encode the current macroblock, and transfer to step (4), if inter-frame prediction is required, proceed to the next step;

(2) Set the search window for the current macroblock, judge whether the reference data in the search window is ready, if yes, encode the current macroblock according to the reference data, and go to step (4), if not, go to the next step step;

(3) Decode the reference code stream to obtain the reference data of a macroblock, and return to step (2);

(4) Judging whether to end encoding, if not, process the next macroblock and return to step (1).

Further, the reference data is I frame type reference data.

Further, the reference data includes reference data of the P frame type, and the reference data of the P frame type is encoded and decoded by referring to the reference frame data of the I frame type.

Further, in step (1), when inter-frame prediction is not required, the process of encoding the current macroblock includes:

Perform intra-frame prediction of the current macroblock according to the reference data obtained after the previous coded macroblock reconstruction processing to obtain an intra-frame prediction value, and then subtract the prediction value from the current macroblock data to obtain the residual of the current macroblock data;

performing forward DCT transformation and quantization on the residual data of the current macroblock;

performing entropy encoding on the residual data after forward DCT transformation and quantization, generating and caching the code stream of the current macroblock;

Perform reverse DCT transformation and dequantization on the residual data after forward DCT transformation and quantization;

The residual data after inverse DCT transformation and dequantization and the prediction data of the best prediction mode are reconstructed to obtain reference data for intra-frame prediction of the next macroblock.

Further, in step (2), the process of preparing the reference data of the current macroblock within the search window includes:

Perform entropy decoding on the cached reference code stream;

Perform inverse DCT transformation and inverse quantization on residual coefficients obtained by entropy decoding;

perform intra-frame prediction according to the intra-frame prediction mode output by entropy decoding;

The residual data after inverse DCT transformation and dequantization and the prediction data after intra-frame prediction are reconstructed to obtain and cache the reference data required in the search window during inter-frame prediction of the current macroblock.

Further, in step (2), after the reference data in the search window is prepared, the process of encoding the current macroblock according to the reference data includes:

Perform inter-frame prediction of the current macroblock according to the reference data, compare the coding costs of intra-frame prediction and inter-frame prediction, select a prediction mode with a smaller coding cost as the best prediction mode of the current macroblock, and obtain the Residual data of the current macroblock;

performing forward DCT transformation and quantization on the residual data of the current macroblock;

performing entropy encoding on the residual data after forward DCT transformation and quantization, generating and caching the code stream of the current macroblock;

Perform reverse DCT transformation and dequantization on the residual data after forward DCT transformation and quantization;

The residual data after inverse DCT transformation and dequantization and the prediction data of the best prediction mode are reconstructed to obtain reference data for intra-frame prediction of the next macroblock.

Compared with the prior art, the H.264 encoder and encoding method provided by the present invention obtain the reference data required by the current macroblock by setting the search window when realizing inter-frame prediction, and the storage space involved only includes The size of the stored reference code stream and part of the reference data is much smaller than the size of a frame of image, thus effectively reducing the storage space of the reference frame required by the H.264 encoder; at the same time, because no off-chip memory is used, the entire H.264 encoding The cost of the device and the power consumption of the system are effectively reduced. Moreover, since the read and write performance of the on-chip storage is generally much better than that of the off-chip storage, when large-scale image encoding is performed, the image quality can be guaranteed while greatly improving the storage capacity. Data throughput.

Description of drawings

Fig. 1 is the structural representation of a kind of H.264 coder of prior art;

Fig. 2 is the structural representation of H.264 coder of the present invention;

Fig. 3 is the simple flow chart of the encoding work of H.264 encoder of the present invention;

FIG. 4 is a schematic structural diagram of an H.264 encoder according to Embodiment 1 of the present invention;

FIG. 5 is a schematic diagram of the encoding sequence and reference relationship of the H.264 encoder according to Embodiment 1 of the present invention;

FIG. 6 is a schematic diagram of the position of the search window corresponding to the macroblock of the H.264 encoder according to Embodiment 1 of the present invention;

FIG. 7 is a schematic diagram of the macroblock correspondence relationship of the encoding/decoding subsystem of the H.264 encoder according to Embodiment 1 of the present invention;

FIG. 8 is a schematic structural diagram of an H.264 encoder according to Embodiment 2 of the present invention;

FIG. 9 is a schematic diagram of the macroblock correspondence relationship of the encoding/decoding subsystem of the H.264 encoder according to Embodiment 2 of the present invention;

FIG. 10 is a schematic structural diagram of an H.264 encoder according to Embodiment 3 of the present invention;

FIG. 11 is a schematic diagram of the encoding sequence and reference relationship of the H.264 encoder according to Embodiment 3 of the present invention;

FIG. 12 is a schematic diagram of macroblock correspondences of the encoding/decoding subsystem of the H.264 encoder according to Embodiment 3 of the present invention.

Detailed ways

Since inter-prediction macroblocks are divided into forward inter-prediction macroblocks and bidirectional inter-prediction macroblocks, there are three main frame types in the H.264 standard: I frame, P frame and B frame. All the macroblocks of the I frame are intra-prediction, the P frame contains the forward interframe prediction macroblock and the intraframe prediction macroblock, and the B frame contains the bidirectional interframe prediction macroblock and the intraframe prediction macroblock. Of course, the realization of P-frame and B-frame encoding requires a reference frame, but I-frame does not.

Therefore, the H.264 encoder and encoding method proposed by the present invention mainly include the following aspects:

Store the encoded bitstream of the reference frame, that is, the reference bitstream;

During inter-frame prediction, set a search window (search window) of a certain size, and search for the best matching block data within the search window;

Obtain the search window data required by the current macroblock by decoding the reference code stream, and store the data;

Considering the independence of decoding the reference code stream, at least one frame in the reference code stream is an I frame.

Therefore, referring to Fig. 2, the H.264 encoder provided by the present invention mainly includes the following four parts:

The encoding subsystem 10 is used to input the data of the current image, perform intra-frame prediction or inter-frame prediction of the current macroblock to obtain the residual data of the current macroblock, and perform DCT transformation, quantization and entropy on the residual data Outputting the code stream of the current macroblock after encoding, wherein, when performing inter-frame prediction on the current macroblock, setting a search window for the current macroblock, and preparing required reference data within the search window;

The code stream buffer module 20 is used to cache the reference code stream, and the reference code stream is the code stream after the reference frame image is encoded, that is, if the current frame will be used as the reference frame of the subsequent frame, then all the macros of the current frame The code stream of the block will be saved until the attribute of the frame as a reference frame is cancelled; if the current frame is not used as the reference frame of the subsequent frame, the code stream of the current macroblock will be sent to the coding system, such as writing SD card, these code streams can be overwritten by subsequent code stream data;

The decoding subsystem 30 is configured to perform entropy decoding, inverse DCT transformation, and inverse quantization on the reference code stream to obtain reference data required in the search window during inter-frame prediction of the current macroblock;

The reference data caching module 40 is configured to cache the reference data obtained by the decoding subsystem, and provide the coding subsystem with the reference data required in the search window during the inter-frame prediction of the current macroblock.

Please refer to Figure 3, the encoding/decoding process of the H.264 encoder provided by the present invention is carried out in units of macroblocks (MB), and while encoding the current MB, the decoding subsystem can also work simultaneously, Preparing reference data for the next macroblock in advance can effectively improve the encoding speed of the entire system. When realizing the inter-frame prediction of the current macroblock, set the search window of the current macroblock, and prepare the reference data needed in the search window, if the reference data is not prepared, then call the decoding subsystem 30, according to the macroblock The reference code stream is further decoded in the raster scan order of the current macroblock to obtain the reference data for inter-frame prediction of the current macroblock.

The H.264 encoder and encoding method proposed by the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

Embodiment one

Please refer to FIG. 4. This embodiment provides an H.264 encoder, which includes the following four parts:

The encoding subsystem 10 is used to input the data of the current image, perform intra-frame prediction or inter-frame prediction of the current macroblock to obtain the residual data of the current macroblock, and perform DCT transformation, quantization and entropy on the residual data Outputting the code stream of the current macroblock after encoding, wherein, when performing inter-frame prediction on the current macroblock, setting a search window for the current macroblock, and preparing required reference data within the search window;

The bitstream buffer module BitstreamBuffer20 is used to cache the reference bitstream, and the reference bitstream is the coded bitstream of the reference frame image;

The decoding subsystem 30 is used to decode the reference code stream, perform inverse DCT transformation and inverse quantization, and obtain the reference data required in the search window during the inter-frame prediction of the current macroblock;

The reference data cache module ReferenceData40 is used to cache the reference data obtained by the decoding subsystem, and provide the encoding subsystem with the reference data required in the search window during the inter-frame prediction of the current macroblock.

Please continue to refer to FIG. 4, in the present embodiment, the coding subsystem 10 includes the following subunits:

The control unit CurrentFrame101 is used to input the current macroblock data, and send a macroblock start signal to other units to notify other units to start the corresponding calculation of the macroblock;

The residual unit is used to perform intra-frame prediction or inter-frame prediction of the current macroblock to obtain the residual data of the current macroblock, wherein, when performing inter-frame prediction on the current macroblock, set the current macroblock Search the window, and prepare the required reference data in the search window, the reference data is obtained by decoding the reference code stream;

A forward DCT transformation and quantization unit DCT/Q103, configured to perform forward DCT transformation and quantization on the residual data;

An entropy encoding unit EntropyEncoder104, configured to perform entropy encoding on the residual data after the forward DCT transformation and quantization, to generate the code stream of the current macroblock;

The inverse DCT transformation and inverse quantization unit IDCT/InvQ105 is used to perform inverse DCT transformation and inverse quantization on the residual data after forward DCT transformation and quantization;

The reconstruction unit 106 is configured to add the residual data after inverse DCT transformation and inverse quantization to the intra-predicted or inter-predicted data of the current macroblock to obtain reconstructed data, which is the next macroblock Intra prediction of blocks provides reference data.

Please continue to refer to Fig. 4, in the present embodiment, the described residual unit includes the following submodules:

The prediction mode selection module ModeDecision102a is used to select an intra prediction or inter prediction mode for the current macroblock according to the size of the encoding cost;

The inter-frame prediction module ME/MC (MotionEstimation/MotionCompensation) 102b is used to set the search window for the current macroblock in the inter-frame prediction mode, and select the reference required by the search window from the reference data cache module data, and perform inter-frame prediction on the current macroblock according to the reference data;

The intra-frame prediction selection module ChooseIntraMode102c is used to select the prediction mode with the smallest coding cost as the best intra-frame prediction of the current macroblock by comparing the coding costs of each intra-frame prediction mode according to the reconstructed data obtained from the previously coded macroblocks model;

An intra prediction module IntraPrediction102d, configured to perform intra prediction on the current macroblock according to the optimal intra prediction mode and the reconstructed data;

The multiplex selection module MUX102e is used to realize the selection of the prediction data output by the prediction mode selection module 102a, the inter-frame prediction module 102b and the intra-frame prediction module 102d;

The calculation module 102f is used for subtracting the input data of the current macroblock from the intra-frame prediction or inter-frame prediction data to obtain the residual data of the current macroblock.

Please continue to refer to FIG. 4. In this embodiment, the decoding subsystem 30 includes the following subunits:

An entropy decoding unit EntropyDecoder301, configured to perform entropy decoding on the reference code stream;

Inverse DCT transformation and inverse quantization unit IDCT/InvQ302, used to perform inverse DCT transformation and inverse quantization on the residual coefficient output by the entropy decoding unit;

An intra-frame prediction unit IntraPrediction303, configured to perform intra-frame prediction according to the intra-frame prediction mode output by the entropy decoding unit and the reconstructed data of the previously decoded macroblock;

The reconstruction unit 304 is configured to add the residual data output by the inverse DCT transformation and inverse quantization unit 302 and the prediction data output by the intra prediction unit 303 to obtain reconstructed data, which is the current macroblock The reference data required in the search window during inter-frame prediction.

In this embodiment, the intra prediction unit 303 of the decoding subsystem 30 and the intra prediction module 102d of the residual unit are two independent intra predictors; the reverse DCT of the decoding subsystem 30 The transform and inverse quantization unit 302 and the inverse DCT transform and inverse quantization unit 105 of the encoding subsystem 10 are two independent inverse DCT transform/dequantizers.

In this embodiment, the decoding subsystem 30 further includes: a filtering unit Deblocking305, configured to filter the reference data obtained by the reconstruction unit 304 before being cached in the reference data caching module, so as to Reduce blockiness. In other embodiments of the present invention, Deblocking 305 may also be omitted.

In this embodiment, BitstreamBuffer20 and ReferenceData40 are preferably SRAM. Among them, the size of BitstreamBuffer20 is related to the size of the encoded image and the selected quantization coefficient. In addition, if B-frame encoding is to be implemented, the current image data must be cached first, and then encoded after the next frame image encoding is completed, which increases the storage space of the encoding system. Therefore, in order to reduce the storage space and reduce the cost of the encoding system, the H.264 encoder of this embodiment is more suitable for encoding I frames and P frames.

In this embodiment, the reference data buffering module ReferenceData40 is a cyclic memory (cyclic buffer) with a storage space of a certain number of macroblocks. When buffering the reference data obtained by the decoding subsystem, the currently obtained reference data will overwrite the The reference data in the circular memory will no longer be used for inter-frame prediction, that is to say, decoding the newly generated data will cover the "old" data; when the basic data unit stored by ReferenceData40 is a macroblock, it needs to store at least The number of macroblocks S can be expressed by the following formula:

S=Ceil(sw_height_by_pix/16)*pic_width_by_mb-(pic_width_by_mb-Ceil(sw_width_by_pix/16))+1

Among them, Ceil(x) represents the smallest integer greater than or equal to x;

sw_height_by_pix is the height of the search window expressed in pixels;

sw_width_by_pix is the width of the search window expressed in pixels;

pic_width_by_mb is the width of the picture expressed in units of macroblocks.

The following example illustrates the specific operation method based on the H.264 encoder shown in Figure 4, assuming that the size of the image to be encoded is 640x480, the image format is Y:Cb:Cr=4:2:0, and the size of the search window is 592x48. Firstly, according to the above formula of the number of macroblocks, it can be obtained that the minimum storage space of ReferenceData40 is 118 macroblocks, but for the convenience of reading and writing addressing, the size of the storage space is set to 120 macroblocks, that is, three complete macroblocks Block bar (MB-line), which is 45Kbyte; in addition, the size of the I-frame stream of a VGA image can generally be controlled within 25Kbyte. To sum up, in order to realize the P-frame encoding of VGA images, the traditional H.264 encoder needs to store at least one reference image, that is, a storage space of 450Kbyte, while using the encoding method of the H.264 encoder of this embodiment, it only needs 70Kbyte storage space. See Figure 5, Figure 6, and Figure 7 for the specific process.

FIG. 5 shows the image encoding sequence of the H.264 encoder in this embodiment and the reference relationship between them. It is worth noting that the P frame will not be limited to only refer to the previous I frame. For example, if the P-24 frame in the figure uses the I-0 frame as a reference, then it must be ensured that the code stream of the I-0 frame is still stored in BitstreamBuffer20 If it is not covered by the code stream of the I-22 frame, of course the price paid is to increase the storage space; in addition, the number of P frames between two I frames can also be arbitrary.

Figure 6 describes the moving process of the search window position. A dotted box in the figure is the search window of the first macroblock (MB0) of the P frame. As the coded macroblocks increase in the order of rasterscan, the position of the search window changes accordingly. Another dotted line in the figure The frame is the search window of MB59 in the P frame. Although the size of the search window can be set freely, in practical applications, the on-chip storage space owned by a coding system is fixed, so the maximum value of the search window can be deduced according to the formula of the number of macroblocks stored in ReferenceData40 above. value.

Figure 7 shows the start-up times of the encoding/decoding subsystems and their relative relationship. The decoding subsystem will start working in advance to decode the macroblocks of the reference frame in the order of rasterscan, and prepare reference data for the current P frame. When the reference data needed for the search window of the first macroblock of the P frame are all ready, the encoding subsystem starts to work, and then the encoding/decoding proceeds simultaneously. When the decoding subsystem reaches MB120, the storage space for reference data in ReferenceData40 is full. At this time, the decoded data of MB120 should be overwritten to the space where MB0 data was originally stored, because the data of MB0 will no longer be used as Reference data, the so-called circular storage. Finally, the decoding subsystem will end early, and if the subsequent frame is still a P frame, the decoding subsystem will decode the reference frame again.

It should be noted that, if the encoding performance of the H.264 encoder in this embodiment is to be improved, on the one hand, the search window can be enlarged; on the other hand, several more frames of reference streams can be stored to form a multi-frame reference. These two points are for finding reference data that better matches the current macroblock. Of course, the cost is to increase the on-chip storage space. As long as the storage space used is smaller than the traditional implementation scheme, it is meaningful to do so.

Embodiment two

For the hardware implementation of the H.264 encoder, it is usually necessary to consider the following four aspects: 1). The area of the hardware implementation; 2). The highest frequency of the working clock; 3). The encoding speed, that is, the processing required for each macroblock 4). Coding performance, that is, the smaller the code stream generated under the same image quality, the better. And these aspects affect each other. For example, to reduce the area used, it is necessary to share a part of the logic unit or reduce the storage space, then the encoding speed or encoding performance will be reduced; if the maximum operating clock frequency is to be increased, then In general, the area will be increased; if the coding performance is to be improved, the computational complexity and storage space will generally be increased so that the predicted value can better match the current macroblock data, which increases the area.

Based on the consideration of the four factors of the hardware implementation of the above-mentioned H.264 encoder, this embodiment provides an H.264 encoder that can relatively save area, in which the encoding subsystem and the decoding subsystem can reuse part of the same logic Blocks, such as functional modules that multiplex inverse DCT/inverse quantization and intra-frame prediction, please refer to Figure 8. The specific structure includes:

The encoding subsystem 10 is used to input the data of the current image, perform intra-frame prediction or inter-frame prediction of the current macroblock to obtain the residual data of the current macroblock, and perform DCT transformation, quantization and entropy on the residual data Outputting the code stream of the current macroblock after encoding, wherein, when performing inter-frame prediction on the current macroblock, setting a search window for the current macroblock, and preparing required reference data within the search window;

The bitstream buffer module BitstreamBuffer20 is used to cache the reference bitstream, the reference bitstream is the encoded bitstream of the reference frame image, that is, if the current frame will be used as the reference frame of the subsequent frame, then all the macroblocks of the current frame The code stream of the current macroblock will be saved until the attribute of the frame as a reference frame is cancelled; if the current frame is not used as the reference frame of the subsequent frame, the code stream of the current macroblock will be sent to the coding system, such as written to SD card, these code streams can be overwritten by subsequent code stream data;

The decoding subsystem 30 is used to perform entropy decoding, inverse DCT transformation and inverse quantization on the reference code stream, so as to obtain the reference data required in the search window during the inter-frame prediction of the current macroblock;

The reference data cache module ReferenceData40 is used to cache the reference data obtained by the decoding subsystem, and provide the encoding subsystem with the reference data required in the search window during the inter-frame prediction of the current macroblock.

Please continue to refer to FIG. 8. In this embodiment, the encoding subsystem 10 includes the following subunits:

The control unit CurrentFrame101 is used to input the current macroblock data, and send a macroblock start signal to other units to notify other units to start the corresponding calculation of the macroblock;

The residual unit is used to perform intra-frame prediction or inter-frame prediction of the current macroblock to obtain the residual data of the current macroblock, wherein, when performing inter-frame prediction on the current macroblock, set the current macroblock Search the window, and prepare the required reference data in the search window, the reference data is obtained by decoding the reference code stream;

A forward DCT transformation and quantization unit DCT/Q103, configured to perform forward DCT transformation and quantization on the residual data;

An entropy encoding unit EntropyEncoder104, configured to perform entropy encoding on the residual data after the forward DCT transformation and quantization, to generate the code stream of the current macroblock;

The inverse DCT transformation and inverse quantization unit IDCT/InvQ105 is used to perform inverse DCT transformation and inverse quantization on the residual data after forward DCT transformation and quantization;

The reconstruction unit 106 is configured to add the residual data Dn after inverse DCT transformation and inverse quantization and the prediction data after intra-frame prediction or inter-frame prediction of the current macroblock to obtain reconstructed data, which is as follows Intra prediction of a macroblock provides reference data.

Please continue to refer to Figure 8, in the present embodiment, the described residual unit includes the following submodules:

The prediction mode selection module ModeDecision102a is used to select an intra prediction or inter prediction mode for the current macroblock according to the size of the encoding cost;

The inter-frame prediction module ME/MC (MotionEstimation/MotionCompensation) 102b is used to set the search window for the current macroblock in the inter-frame prediction mode, and select the reference required by the search window from the reference data cache module data, and perform inter-frame prediction on the current macroblock according to the reference data;

The intra-frame prediction selection module ChooseIntraMode102c is used to select the best intra-frame prediction mode of the current macroblock by comparing the encoding costs of each intra-frame prediction mode according to the reconstructed data obtained from the previously encoded macroblocks;

An intra prediction module IntraPrediction102d, configured to perform intra prediction on the current macroblock according to the optimal intra prediction mode and the reconstructed data;

The multiplex selection module MUX102e is used to realize the selection of the prediction data output by the prediction mode selection module 102a, the inter-frame prediction module 102b and the intra-frame prediction module 102d;

The calculation module 102f is used for subtracting the input data of the current macroblock from the intra-frame prediction or inter-frame prediction data to obtain the residual data of the current macroblock.

Please continue to refer to FIG. 8. In this embodiment, the decoding subsystem 30 includes the following subunits:

An entropy decoding unit EntropyDecoder301, configured to perform entropy decoding on the reference code stream;

The inverse DCT transformation and inverse quantization unit IDCT/InvQ105, that is, the inverse DCT transformation and inverse quantization unit 105 of the encoding subsystem 10 is used to perform inverse DCT transformation and inverse quantization on the residual coefficient output by the entropy decoding unit ;

The intra-frame prediction unit IntraPrediction102d, that is, the intra-frame prediction module 102d of the residual unit, is used to perform intra-frame prediction according to the intra-frame prediction mode output by the entropy decoding unit 301 and the previously decoded macroblock reconstruction data. predict;

The reconstruction unit 304 is used to add the residual data output by the inverse DCT transformation and inverse quantization unit 105 and the prediction data output by the intra prediction module 102d to obtain reconstructed data, which is used as the macroblock inter prediction search The reference data required in the window.

In this embodiment, the decoding subsystem 30 further includes: a filtering unit Deblocking305, configured to filter the reference data obtained by the reconstruction unit 304 before being cached in the reference data caching module, so as to Reduce blockiness. In other embodiments of the present invention, Deblocking 305 may also be omitted.

Please refer to Fig. 9, since the encoding subsystem and the decoding subsystem in the H.264 encoder of this embodiment reuse part of the logic, then the encoding subsystem and the decoding subsystem cannot work at the same time, and only time-sharing can be used for encoding and decoding. For decoding, it takes more clock cycles to process one macroblock data, so the encoding speed is reduced compared with the first embodiment.

Embodiment three

Similarly, based on the four considerations of the hardware implementation of the H.264 encoder described in Embodiment 2, this embodiment provides an H.264 encoder that improves encoding performance by storing several more frames of reference streams to form a multi-frame reference. .264 encoder, please refer to Figure 10, the specific structure includes:

The encoding subsystem 10 is used to input the data of the current image, perform intra-frame prediction or inter-frame prediction of the current macroblock to obtain the residual data of the current macroblock, and perform DCT transformation, quantization and entropy on the residual data Outputting the code stream of the current macroblock after encoding, wherein, when performing inter-frame prediction on the current macroblock, setting a search window for the current macroblock, and preparing required reference data within the search window;

The bitstream buffer module BitstreamBuffer20 is used to cache the reference bitstream, the reference bitstream is the encoded bitstream of the reference frame image, that is, if the current frame will be used as the reference frame of the subsequent frame, then all the macroblocks of the current frame The code stream of the current macroblock will be saved until the attribute of the frame as a reference frame is cancelled; if the current frame is not used as the reference frame of the subsequent frame, the code stream of the current macroblock will be sent to the coding system, such as written to SD card, these code streams can be overwritten by subsequent code stream data;

The decoding subsystem 30 is used to perform entropy decoding, inverse DCT transformation and inverse quantization on the cached reference code stream, so as to obtain the reference data required in the search window during the inter-frame prediction of the current macroblock;

The reference data cache module ReferenceData40 is used to cache the reference data obtained by the decoding subsystem, and provide the encoding subsystem with the reference data required in the search window during the inter-frame prediction of the current macroblock.

Please continue to refer to Figure 10. In this embodiment, the coding subsystem 10 includes the following subunits:

The control unit CurrentFrame101 is used to input the current macroblock data, and send a macroblock start signal to other units to notify other units to start the corresponding calculation of the macroblock;

The residual unit is used to perform intra-frame prediction or inter-frame prediction of the current macroblock to obtain the residual data of the current macroblock, wherein, when performing inter-frame prediction on the current macroblock, set the current macroblock Search the window, and prepare the required reference data in the search window, the reference data is obtained by decoding the reference code stream;

A forward DCT transformation and quantization unit DCT/Q103, configured to perform forward DCT transformation and quantization on the residual data;

An entropy encoding unit EntropyEncoder104, configured to perform entropy encoding on the residual data after the forward DCT transformation and quantization, to generate the code stream of the current macroblock;

The inverse DCT transformation and inverse quantization unit IDCT/InvQ105 is used to perform inverse DCT transformation and inverse quantization on the residual data after forward DCT transformation and quantization;

The reconstruction unit 106 is configured to add the residual data Dn after inverse DCT transformation and inverse quantization and the prediction data after intra-frame prediction or inter-frame prediction of the current macroblock to obtain Reference data for intra prediction.

Please continue to refer to Figure 10, in this embodiment, the described residual unit includes the following submodules:

The prediction mode selection module ModeDecision102a is used to select an intra prediction or inter prediction mode for the current macroblock according to the size of the encoding cost;

The inter-frame prediction module ME/MC (MotionEstimation/MotionCompensation) 102b is used to set the search window for the current macroblock in the inter-frame prediction mode, and select the reference required by the search window from the reference data cache module data, and perform inter-frame prediction on the current macroblock according to the reference data;

The intra-frame prediction selection module ChooseIntraMode102c is used to select the prediction mode with the smallest coding cost as the best intra-frame prediction of the current macroblock by comparing the coding costs of each intra-frame prediction mode according to the reconstructed data obtained from the previously coded macroblocks model;

An intra prediction module IntraPrediction102d, configured to perform intra prediction on the current macroblock according to the optimal intra prediction mode and the reconstructed data;

The multiplex selection module MUX102e is used to realize the selection of the prediction data output by the prediction mode selection module 102a, the inter-frame prediction module 102b and the intra-frame prediction module 102d;

The calculation module 102f is used for subtracting the input data of the current macroblock from the intra-frame prediction or inter-frame prediction data to obtain the residual data of the current macroblock.

Please continue to refer to Fig. 10, in the present embodiment, because there is P frame wherein as reference data, so will add a set of logic that can decode P frame in decoding subsystem, therefore, described decoding subsystem 30 includes The P frame decoding system for providing the reference data of the P frame type and the I frame decoding system for the reference data of the I frame type.

Wherein, please continue to refer to Figure 10, the I frame decoding system includes:

The first entropy decoding unit EntropyDecoder301b is used to perform entropy decoding on the I-frame reference code stream in the reference code stream;

The first inverse DCT transformation and inverse quantization unit IDCT/InvQ302b, configured to perform inverse DCT transformation and inverse quantization on the residual coefficient output by the first entropy decoding unit 301b;

The first intra-frame prediction unit IntraPrediction303b is configured to perform intra-frame prediction according to the intra-frame prediction mode output by the first entropy decoding unit 301b;

The first reconstruction unit 304b is configured to add the residual data output by the first inverse DCT transform and inverse quantization unit 302b and the prediction data output by the first intra-frame prediction unit 303b to obtain the I frame reference The code stream provides the reference data required in the search window for the inter-frame prediction of the current macroblock, and the reference data can be used for inter-frame prediction when the P-frame reference code stream is decoded.

The first filtering unit Deblocking 305b is configured to filter the reference data of the first reconstruction unit 304b.

The P frame decoding system includes:

The second entropy decoding unit EntropyDecoder301a is used to perform entropy decoding on the P frame reference code stream in the reference code stream;

The second inverse DCT transformation and inverse quantization unit IDCT/InvQ302a, configured to perform inverse DCT transformation and inverse quantization on the residual coefficient output by the second entropy decoding unit 301a;

The second intra-frame prediction unit IntraPrediction303a is configured to perform intra-frame prediction according to the intra-frame prediction mode output by the second entropy decoding unit 301a;

The second inter-frame prediction unit MC306 is configured to perform inter-frame prediction according to the reconstructed data of the first reconstruction unit 304b and the motion vector output by the second entropy decoding unit 301a;

a multiplexer MUX307, configured to select the prediction data output by the second intra prediction unit IntraPrediction303a and the second inter prediction unit MC306;

The second reconstruction unit 304a is configured to perform the residual data output by the second inverse DCT transform and inverse quantization unit 302a and the prediction data processed by the second intra prediction unit 303a or the second inter prediction unit 306 Add up to obtain the reconstructed data of the P-frame reference code stream, which is used for the reference data required in the search window for the inter-frame prediction of the current macroblock, and if the next macroblock of the reference code stream is a frame Intra-prediction, then the reconstructed data is also used as reference data for intra-frame prediction during decoding;

The second filtering unit Deblocking 305a is configured to filter the reference data of the second reconstruction unit 304a.

Please continue to refer to FIG. 10 , in this embodiment, the reference data cache module includes an I frame cache module 401 for caching reference data of an I frame type, a P frame cache module 402 for caching a P frame type, and a P frame cache module 402 for selecting an I frame type or A selector 403 for reference data of P frame type.

In other embodiments of the present invention, Deblocking 305a, 305b can also be omitted, and the principle of Embodiment 2 can also be referred to, and the modules or/or units with the same function in the encoding subsystem and the decoding subsystem can be partially multiplexed, and after multiplexing Only affects the relative relationship between encoding/decoding subsystem operation.

FIG. 11 shows the encoding sequence and reference relationship of the H.264 encoder in this embodiment. It is worth noting that the P' frame as the reference frame must be after the I frame, and the P' frame must use the I frame as a reference. In addition, the subsequent P frame can use either the P' frame as a reference, or the reference frame (ie, the I frame) of the P' frame as a reference.

Figure 12 depicts the relative relationship between the encoding/decoding subsystem operation. The figure still assumes that the encoded image size is 640x480 and the search window size is 592x48. It can be seen from Figure 12 that if the current frame needs to refer to the P' frame, then the reference frame of the P' frame (that is, the I frame) must be decoded first, and then the P' frame can be decoded, and the current P' frame can only be decoded after the above two steps are completed. frames are encoded.

Although the H.264 encoder in this embodiment increases the implementation complexity of the system and increases the on-chip storage space, since it provides multi-frame reference, it improves the performance of the encoding system and can be used as a choice for practical applications.

In addition, the present invention also provides a kind of H.264 coding method, comprises the following steps:

(1) Determine whether the current macroblock needs to perform inter-frame prediction, if no inter-frame prediction is required, encode the current macroblock, and transfer to step (4), if inter-frame prediction is required, proceed to the next step;

(2) Set the search window for the current macroblock, judge whether the reference data in the search window is ready, if yes, encode the current macroblock according to the reference data, and go to step (4), if not, go to the next step step;

(3) Decode the reference code stream to obtain the reference data of a macroblock, and return to step (2);

(4) Judging whether to end encoding, if not, process the next macroblock and return to step (1).

The reference data may be reference data of I frame type, and may also include reference data of P frame type, and the reference data of P frame type is obtained by encoding and decoding with reference to reference frame data of I frame type.

In step (1), when inter-frame prediction is not required, the process of encoding the current macroblock includes:

Perform intra-frame prediction of the current macroblock according to the reference data obtained after the previous coded macroblock reconstruction processing to obtain an intra-frame prediction value, and then subtract the prediction value from the current macroblock data to obtain the residual of the current macroblock data;

performing forward DCT transformation and quantization on the residual data of the current macroblock;

performing entropy encoding on the residual data after forward DCT transformation and quantization, generating and caching the code stream of the current macroblock;

Perform reverse DCT transformation and dequantization on the residual data after forward DCT transformation and quantization;

The residual data after inverse DCT transformation and dequantization and the prediction data of the best prediction mode are reconstructed to obtain reference data for intra-frame prediction of the next macroblock.

In step (2), the process of preparing the reference data of the current macroblock within the search window includes:

Perform entropy decoding on the cached reference code stream;

Perform inverse DCT transformation and inverse quantization on residual coefficients obtained by entropy decoding;

perform intra-frame prediction according to the intra-frame prediction mode output by entropy decoding;

The residual data after inverse DCT transformation and dequantization and the prediction data after intra-frame prediction are reconstructed to obtain and cache the reference data required in the search window during inter-frame prediction of the current macroblock.

In step (2), after the reference data in the search window is ready, the process of encoding the current macroblock according to the reference data includes:

Perform inter-frame prediction of the current macroblock according to the reference data, compare the coding costs of intra-frame prediction and inter-frame prediction, select a prediction mode with a smaller coding cost as the best prediction mode of the current macroblock, and obtain the Residual data of the current macroblock;

performing forward DCT transformation and quantization on the residual data of the current macroblock;

performing entropy encoding on the residual data after forward DCT transformation and quantization, generating and caching the code stream of the current macroblock;

Perform reverse DCT transformation and dequantization on the residual data after forward DCT transformation and quantization;

The residual data after inverse DCT transformation and dequantization and the prediction data of the best prediction mode are reconstructed to obtain reference data for intra-frame prediction of the next macroblock.

It can be seen from the above that the H.264 encoder and encoding method proposed in this embodiment do not store all the original data of the reference frame, but refine it to the macroblock level to store the reference data required by the current macroblock. The data is obtained by decoding the reference code stream, so that the storage space of the reference code stream plus some reference data replaces the storage space of the entire reference frame. Since the stored reference data must contain the data of the entire search window, the storage space of the reference data is related to the size of the search window, and according to the image compression performance, the size of the search window can be set freely, but generally it is much smaller than the size of a frame of image size, which reduces the storage space required by the H.264 encoder. Therefore, according to the above method, H.264 encoding can be performed using a smaller on-chip SRAM. The specific hardware implementation scheme is shown in FIG. 3 . Since no off-chip memory is used, the cost of the entire coding system is reduced, and the power consumption of the system is also reduced. Moreover, the read and write performance of the on-chip SRAM is generally much better than that of the off-chip SDRAM memory, so the data throughput rate of the memory will not become a problem when encoding large-size images.

In summary, the H.264 encoder and encoding method provided by the present invention obtain the reference data required by the current macroblock by setting the search window when realizing inter-frame prediction, and the storage space involved only includes the stored The code stream and some reference data are far smaller than the size of a frame image, thus effectively reducing the storage space of the reference frame required by the H.264 encoder; at the same time, because no off-chip memory is used, the cost of the entire H.264 encoder And the power consumption of the system is effectively reduced. Moreover, since the read and write performance of on-chip storage is generally much better than that of off-chip storage, when encoding large-size images, the storage data throughput can be greatly improved while ensuring image quality. .

Obviously, those skilled in the art can make various changes and modifications to the invention without departing from the spirit and scope of the invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention also intends to include these modifications and variations.

Claims (12)

1. a H.264 encoder, is characterized in that, comprising:

Code-subsystem, for inputting the data of present image, carry out the infra-frame prediction of current macro or inter prediction to obtain the residual error data of described current macro, the code stream of described current macro is exported after dct transform and quantification and entropy code are carried out to described residual error data, wherein, when carrying out inter prediction to current macro, the search window of setting current macro, and in search window, get out the reference data of needs;

Code stream cache module, for buffer memory with reference to code stream, described reference code stream is the code stream of reference frame image after coding;

Decoding sub-system, for carrying out entropy decoding and oppositely dct transform and inverse quantization to described reference code stream, obtains reference data required when described code-subsystem carries out inter prediction;

Reference data cache module for the reference data that decoding sub-system described in buffer memory obtains, and is reference data required in search window when described code-subsystem provides the inter prediction of current macro;

Wherein, described reference data cache module is circulating memory, when the reference data that decoding sub-system described in buffer memory obtains, the reference data of current acquisition covers in described circulating memory the reference data that can not re-use when carrying out inter prediction, described reference data cache module is provided with the memory space of certain macroblock number, and the computing formula of its macroblock number S stored is:

S＝Ceil(sw_height_by_pix/16)*pic_width_by_mb-(pic_width_by_mb-Ceil(sw_width_by_pix/16))+1

Wherein, Ceil (x) represents the smallest positive integral being more than or equal to x;

Sw_height_by_pix is the height of the search window represented in units of pixel;

Sw_width_by_pix is the width of the search window represented in units of pixel;

Pic_width_by_mb is the width of the image represented in units of macro block.

2. H.264 encoder as claimed in claim 1, it is characterized in that, described code-subsystem comprises:

Control unit, for inputting current macro data, and provides a macro block enabling signal and is distributed to other unit to notify that other unit starting macro blocks calculate accordingly;

Ask residual error unit, for carrying out infra-frame prediction or the inter prediction of current macro, obtain the residual error data of described current macro, wherein, when carrying out inter prediction to current macro, the search window of setting current macro, and required reference data is got out in described search window, this reference data obtains by decoding to reference code stream;

Forward dct transform and quantifying unit, for carrying out forward dct transform and quantification to described residual error data;

Entropy code unit, for described forward dct transform and the residual error data after quantizing are carried out entropy code, generates the code stream of described current macro;

Reverse dct transform and inverse quantization unit, for carrying out reverse dct transform and inverse quantization to forward dct transform and the residual error data after quantizing;

Reconfiguration unit, for being added the prediction data after the infra-frame prediction of the residual error data after reverse dct transform and inverse quantization and described current macro or inter prediction, obtains reconstruct data, for the infra-frame prediction of next macro block provides reference data.

3. H.264 encoder as claimed in claim 2, is characterized in that, described in ask residual error unit to comprise:

Predictive mode selects module, for according to the size of Coding cost being current macro selection infra-frame prediction or inter-frame forecast mode;

Inter prediction module, under inter-frame forecast mode, is current macro setting search window, and from described reference data cache module, select the reference data of described search window needs, and carries out inter prediction according to described reference data to current macro;

Intra prediction mode selection module, for the reconstruct data obtained according to previously encoded macro block, the predictive mode selecting Coding cost minimum is as the optimum frame inner estimation mode of current macro;

Intra-framed prediction module, for carrying out infra-frame prediction according to described optimum frame inner estimation mode and described reconstruct data to current macro;

Multi-path choice module, for realizing the selection of the prediction data that predictive mode selects module, intra-framed prediction module and Inter prediction module to export.

4. H.264 encoder as claimed in claim 2, it is characterized in that, described decoding sub-system comprises:

Entropy decoding unit, for carrying out entropy decoding to reference code stream;

Reverse dct transform and inverse quantization unit, carry out reverse dct transform and inverse quantization for the residual error coefficient exported described entropy decoding unit;

Intraprediction unit, for carrying out infra-frame prediction according to the described intra prediction mode of entropy decoding unit output and the reconstruct data of previous decoded macro block;

Reconfiguration unit, for being added the prediction data of the reverse dct transform of described decoding sub-system and the residual error data of inverse quantization unit output and intraprediction unit output, obtain reconstruct data, as reference data required in search window during macro block inter prediction.

5. H.264 encoder as claimed in claim 4, is characterized in that, the intraprediction unit of described decoding sub-system and the described intra-framed prediction module of residual error unit of asking are multiplexing same intra predictor generator or two independently intra predictor generator.

6. H.264 encoder as claimed in claim 4, it is characterized in that, the reverse dct transform of the reverse dct transform of described decoding sub-system and inverse quantization unit and described code-subsystem and inverse quantization unit are multiplexing same reverse dct transform/inverse DCT or two independently reverse dct transform/inverse DCT.

7. H.264 encoder as claimed in claim 4, it is characterized in that, described decoding sub-system also comprises: filter unit, and the reference data for obtaining at described decoding sub-system carries out filtering to described reference data before being cached to described reference data cache module.

8. H.264 encoder as claimed in claim 1, it is characterized in that, described decoding sub-system comprises the I frame decoding system of the reference data for providing I frame type, and described I frame decoding system comprises:

First entropy decoding unit, for carrying out entropy decoding to described with reference to the I frame reference code stream in code stream;

First reverse dct transform and inverse quantization unit, carries out reverse dct transform and inverse quantization for the residual error coefficient exported described first entropy decoding unit;

First intraprediction unit, carries out infra-frame prediction for the intra prediction mode exported according to described first entropy decoding unit;

First reconfiguration unit, prediction data for exporting residual error data and first intra-framed prediction module of the described first reverse dct transform and inverse quantization unit output is added, obtain the reconstruct data of described I frame with reference to code stream, for reference data required in its search window when the decoding of P frame with reference to code stream or the inter prediction as current macro;

First filter unit, for carrying out filtering to the reference data of the first reconfiguration unit.

9. H.264 encoder as claimed in claim 8, it is characterized in that, described decoding sub-system also comprises the P frame decoding system of the reference data for providing P frame type, and described P frame decoding system comprises:

Second entropy decoding unit, for carrying out entropy decoding to described with reference to the P frame reference code stream in code stream;

Second reverse dct transform and inverse quantization unit, for carrying out reverse dct transform and inverse quantization to the residual error coefficient of described second entropy decoding unit;

Second intraprediction unit, carries out infra-frame prediction for the intra prediction mode exported according to described second entropy decoding unit;

Second inter prediction unit, carries out inter prediction for the motion vector exported according to reconstruct data and the described second entropy decoding unit of the first described reconfiguration unit;

Second reconfiguration unit, for being added the residual error data of the described second reverse dct transform and inverse quantization unit output and the prediction data of the second intraprediction unit or the output of the second inter prediction unit, obtain described P frame with reference to the reconstruct data of code stream, reference data required in its search window during inter prediction for described current macro;

Second filter unit, for carrying out filtering to the reference data of the second reconfiguration unit.

10. the H.264 coding method using the H.264 encoder according to any one of claim 1 ~ 9 to realize, is characterized in that, comprise the steps:

(1) judging that current macro is the need of carrying out inter prediction, if do not need inter prediction, then encodes to current macro, proceeding to step (4), if need inter prediction, then entering next step;

(2) be current macro setting search window, judge whether the reference data in search window is ready to, if so, then according to reference data, current macro is encoded, proceed to step (4), if not, then enter next step;

(3) reference code stream is decoded, obtain the reference data of a macro block, return step (2);

(4) judge whether to terminate coding, if do not terminated, then process next macro block and return step (1).

11. H.264 coding methods as claimed in claim 10, it is characterized in that, described reference data is the reference data of I frame type.

12. H.264 coding methods as claimed in claim 10, it is characterized in that, described reference data comprises the reference data of P frame type, the reference data of described P frame type be with the reference frame data of I frame type for reference to encoding, decoding obtains.