JP2919986B2 - Image signal decoding device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention
The present invention is applied to a decoding device that decodes a transmitted or recorded image.

[0002]

2. Description of the Related Art When an image signal picked up by a solid-state image pickup device represented by a CCD or the like is stored as digital data in a storage device such as a memory card, a magnetic disk or a magnetic tape, the amount of data becomes enormous. Therefore, in order to record in a limited storage capacity range,
It is necessary to perform some highly efficient compression on the obtained image signal data.

As a highly efficient image data compression method,
An encoding method using orthogonal transform encoding is widely known, and an example of this method will be described below with reference to FIG.

First, the input image data (f) is divided into blocks of a predetermined size to obtain (fb), and two-dimensional DCT (discrete cosine transform) is performed for each of the divided blocks as orthogonal transform ( F). Next, linear quantization according to each frequency component is performed, and this quantized value (F
_Q ) is subjected to Huffman coding as variable-length coding, and the result is transmitted or recorded as compressed data (C). At this time, the quantization width of the linear quantization is determined by preparing a quantization matrix representing a relative quantization characteristic taking into account the visual characteristic for each frequency component, and multiplying this quantization matrix by a constant. Has been determined. On the other hand, when reproducing image data from compressed data,
The quantized value (F _Q ) of the transform coefficient is obtained by decoding the variable length code (C), but it is impossible to obtain the true value (F) before the quantization from this value. The result obtained by the quantization is (F ′) including an error. This error is a so-called quantization error, which causes deterioration of the image quality of the reproduced image (f ').

More specifically, first, FIG.
As shown in (a), one frame of image data is divided into blocks of a predetermined size (for example, blocks A, B, C,... Formed of 8 × 8 pixels), and each of the divided blocks is orthogonal. Two-dimensional DCT is performed as conversion, and the data is sequentially stored on an 8 × 8 matrix.

When viewing image data on a two-dimensional plane,
It has a spatial frequency which is frequency information based on the distribution of density information. Therefore, by performing the above DCT,
As shown in FIG. 3B, the image data is converted into a DC component DC and an AC component AC, and data indicating the value of the DC component DC at the origin position (0, 0 position) on the 8 × 8 matrix. , And at the 0,7 position, the AC component AC in the horizontal axis direction.
Data indicating the maximum frequency value of AC component AC, data indicating the maximum frequency value of AC component AC in the vertical axis direction at position 7,0, and maximum frequency value of AC component AC in the oblique direction at position 7,7. Data indicating a value is stored, and at the intermediate position, frequency data in a direction associated with each coordinate position is stored in such a manner that data having a frequency higher than the origin side appears.

Next, linear quantization corresponding to each frequency component is performed by dividing the data stored at each coordinate position in the matrix by the quantization width for each frequency component.
Huffman coding is performed on the quantized value as variable length coding. At this time, the difference value between the DC component DC and the DC component of the neighboring block is Huffman-coded.
With respect to the AC component AC, a scan from a low frequency component to a high frequency component called a zigzag scan is performed, and a continuous number of invalid (value is “0”) components (zero run number) and a succeeding valid component The data is subjected to two-dimensional Huffman encoding of the value to obtain data. In this method, the compression ratio is generally controlled by changing the quantization width of the quantization. The higher the compression ratio, the larger the quantization width, and therefore the larger the quantization error and the larger the reproduction image. Image quality degradation becomes noticeable. The quantization error of the transform coefficient tends to appear as so-called block distortion in which a discontinuity occurs at a block boundary portion in a reproduced image. Since the block distortion is visually conspicuous, the subjective impression deteriorates even if the S / N ratio is good. Therefore, the image reproduced by the decoder is subjected to low-pass (low-pass) processing as distortion removal processing. ) A way to filter has been devised. This postfilter can remove distortion relatively well, but if the image contains edges etc., they will be blurred, and conversely if the degree of low pass is reduced to reduce blurring There was a problem that block distortion could not be completely removed. Therefore, in order to solve this problem, the extent to which the transform coefficients are transmitted is determined, and the coefficients within the range are quantized, coded and transmitted. On the decoder side, there is a method in which the information is used to locally change the degree of the post-filter.

[0008]

However, in the above-mentioned method, the encoder sends a signal indicating the range in which the transform coefficient has been sent, and changes the degree of the filter in accordance with the signal. There is a problem that information is wasted as much as the signal indicating the range must be transmitted, and therefore the compression ratio cannot be increased much.

The image signal decoding apparatus of the present invention has been made in view of such a problem, and an object thereof is to adaptively block a highly compressed image with a simple circuit. An object of the present invention is to provide an image data decoding device capable of performing distortion removal processing.

[0010]

In order to solve the above-mentioned problems, an image signal decoding apparatus according to the present invention divides image data into blocks, and performs an orthogonal transform for each of the divided blocks. Variable-length code decoding means for quantizing the transformed output by a quantization means, and thereafter providing the quantized output to a variable-length coding means to decode the variable-length-encoded and compressed image data; An inverse orthogonal transforming means; a distortion removing means for performing a distortion removing process on a transform output from the inverse orthogonal transforming means; and a code amount of each of the blocks or the number of significant coefficients of transform coefficients of each of the blocks. Or a distortion removal function that changes the distortion removal characteristic of the distortion removal means based on any of the sums of weights determined in advance by respective sequences of significant coefficients of each block. It is provided with a determination unit.

[0011]

That is, in the present invention, a distortion removing circuit for performing a distortion removing process on output data obtained by inversely quantizing and inverse orthogonal transforming the decoded output from the variable length code decoding means. Of the code amount of each block, the number of significant coefficients of transform coefficients of each block, or the sum of weights determined in advance by respective sequences of significant coefficients of transform coefficients of each block. It is changed based on crab.

[0012]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS First, the basic concept of the present invention will be described.

In general, the degree of conspicuousness of block distortion changes depending on the spatial frequency of a nearby image. That is,
When block distortion occurs in a portion having components up to a high spatial frequency having a fine structure, the block distortion is not so noticeable. Conversely, when block distortion occurs in a portion having only a low spatial frequency component that changes relatively slowly, the block distortion becomes more conspicuous.

On the other hand, block distortion is caused by discontinuity at block boundaries and has components up to very high spatial frequencies. Therefore, the block distortion can be made inconspicuous by removing a spatial frequency component higher than the frequency of the image near the distortion. Therefore, it is necessary to determine the extent to which the frequency component is included in each block, and adaptively change the distortion removal characteristic based on the value to remove the block distortion without blurring the image. Can be.

[0015] In the encoding to which the present invention is applied, as the compression ratio is increased, the quantization width increases, and the probability that the coefficient is quantized to zero increases. In particular, since the high-frequency components generally have low power, most of them are quantized to 0, and few remain as significant coefficients. Therefore, when two-dimensional Huffman coding or the like of the zero run number and the subsequent significant coefficient value is used, the generated code amount of a block having a small high-frequency component is reduced, and a block having a significant coefficient up to the high-frequency component is generated. Tends to increase the generated code amount.
In other words, it can be seen that when the code amount in the block is large, the coefficient has a value up to a high frequency coefficient. Therefore, in the present invention, the amount of frequency components contained in each block is determined by using the code amount of each block, the number of non-zero transform coefficients, and the like. Is adaptively changed. The method for determining the distortion removal characteristic will be described below with reference to FIG.

As described above, the band in which a significant coefficient exists is, for example, the highest frequency (f
If it is found that the frequency band consists of up to half the frequency component of (max), it means that the band has a band up to (fmax / 2). In other words, it is only necessary to perform a band cut in which the hatched portion in FIG. This is represented by the following equation: H = F × G (1) Here, equation (1) represents filtering in the frequency domain, F, G, and H are coefficients on the Fourier plane of the data, the filter, and the processing result, respectively.
That is, the band transmission characteristics as shown in FIG. However, when performing filtering in the spatial frequency plane, it is not possible to perform processing while changing the characteristics of the filter. There is also a problem that the influence of blocking must be considered. Therefore, this filtering is realized by convolution in the real space, and the convolved coefficient is changed.

H = f * g (2) In equation (2), f, g, and h are the inverse Fourier transform results of F, G, and H, respectively, and the filtering of equation (1) is processed in real space. And the convolution of data f and g as shown in equation (2). Since the kernel size of g is finite, it is impossible to obtain a sharp cutoff characteristic as shown in FIG. 6, but there is no problem in practical use, and the filter coefficient and the kernel size can be determined arbitrarily. Determined in consideration of cutoff characteristics. Hereinafter, embodiments of the present invention will be described with reference to the drawings.

FIG. 1 shows a first embodiment of an image signal decoding apparatus according to the present invention.
FIG. 3 is a block diagram of an embodiment of FIG. In the image signal decoding apparatus of this embodiment, first, the compressed image data transmitted or recorded by the variable length code decoding circuit 11 is received, and the variable length code of the compressed image data is decoded. This decoded output is inversely quantized by the inverse quantization circuit 12. The result of the inverse quantization is sent to the inverse orthogonal transform circuit 13 to obtain an image signal in the real space. For this result, the distortion removal circuit 1
In step 4, distortion removal processing is performed. The method of determining the distortion removal characteristic from the image signal in the real space is as follows. The code amount in each block is monitored by the intra-block code amount counting circuit 15 for each block. The distortion elimination characteristic is determined by the distortion elimination characteristic determination circuit 16. The distortion removal characteristic determination circuit 16 uses a low-pass filter characteristic that preserves a wide frequency band for the block having a large code amount in the block, and conversely, performs a distortion removal characteristic such that the frequency band is not largely preserved for a block having a small code amount. Processing is performed.

Now, the orthogonal transform coefficient of the block of interest given to the inverse orthogonal transform circuit 13 is calculated as 8 pixels.times.
It is assumed that a DCT coefficient of eight pixels is used, and a coefficient of a 4 × 4 portion indicated by hatching is a non-zero, so-called significant coefficient. Similarly, assuming that the DCT coefficient of the adjacent block is not zero in the 3 × 3 portion as shown in FIG. 2B, the code amount obtained by the intra-block code amount counting circuit 15 is larger for the target block than for the target block. More than the next block. At this time, it can be seen that the target block is wider than the frequency band. Therefore, the filter characteristics for the block of interest having a large code amount may be such that only high frequencies are cut in both the horizontal and vertical directions. Conversely, the adjacent block may have such a characteristic as to cut to a lower frequency component.

At this time, since the intra-block code amount counting circuit 15 detects the block boundary in the compressed code data and counts the code amount included in the block boundary, the processing is very simple and the circuit scale can be reduced. it can.

By doing so, it is possible to perform filtering so that the band of information recorded in block units is hardly lost. That is, since a block with a small code amount has only a low spatial frequency component, strong low-pass filtering for averaging over a wide range is performed, and a block with a large code amount includes a relatively high spatial frequency component. So by performing weak low-pass filtering that does not blur much,
Low-pass filtering that does not blur the structure in the block can be realized.

Hereinafter, a second embodiment of the present invention will be described.
According to this embodiment, the frequency band of the block is obtained using the number of the significant coefficients. In this method, it is determined whether or not the coefficient decoded by the variable length code decoding circuit 11 is a significant coefficient, and the number of significant coefficients in each block is counted. According to this method, the processing is more complicated than in the above-described method, but since the information of the transform coefficient is directly used, it is possible to adopt a more preferable distortion removal characteristic for the block. Here, the significant coefficient may be a coefficient that has not become zero due to quantization, or may be a coefficient determined by comparing an absolute value of a transform coefficient with a predetermined threshold.

Hereinafter, a third embodiment of the present invention will be described.
According to this embodiment, the distortion removal characteristics are determined using the sum of the weights determined in advance by the respective sequences of the significant coefficients. This method will be described with reference to FIG. 4. If the coefficient decoded by the variable-length code data circuit 31 is determined to be zero by the coefficient determination circuit 35, the weight addition circuit 36 determines the frequency of the non-zero coefficient. The corresponding weights are added, and the sum is used to determine the distortion removal characteristic. According to this method, assuming that weights as shown in FIG. 5A are used, if the weight addition circuit 36 is configured to have an 8-bit register and stop the addition when overflow occurs, the register Can be classified into four types depending on whether the value belongs to 0.9 or less, 169 or less, or more. Since these four types correspond to the frequency bands of the blocks as they are, the distortion removal characteristics can be determined based on the results, and the frequency bands of the blocks can be obtained more accurately than in the above-described embodiment.

When adding weights, it is also possible to use a logical sum instead of an arithmetic sum. Each sequence of transform coefficients is divided into 25 regions as shown in FIG. 5 (b), and the weight of each region is divided into upper bits corresponding to the vertical frequency and lower bits corresponding to the horizontal frequency. Each is represented by 4 bits. For example, assuming that there are significant coefficients at positions 4 and 6 indicated by a star in the figure, the logical sum of the register value and 00100100 is obtained. That is, each bit serves as a flag as to whether or not there is significant data in the frequency band corresponding to that bit, so that the band can be obtained separately in both the vertical and horizontal directions. Therefore, it is possible to select different vertical and horizontal distortion removal characteristics, and to perform optimal distortion removal processing respectively. Although this process is more complicated than that of the above-described method, it is possible to select a preferable distortion removal characteristic for each block in each direction, so that a strong edge exists in the vertical or horizontal direction. This is convenient in some cases.

Therefore, according to the present invention, a convolution low-pass filter for preserving a band of significant data of transform coefficients obtained by decoding compressed data is adaptively applied to each block. This makes it possible to remove distortion without blurring the structure in each block. Also, since the distortion removal characteristic is determined using only the image data that is an intermediate result of the decoding process, there is no need to add information for distortion removal on the encoder side.
Naturally, processing for detecting the presence or absence of edges in an image or detecting block distortion is not required, so that it can be realized with a very simple circuit configuration, and the processing content counts the code amount and significant coefficient in the block. Therefore, the processing time can be shortened.

When the present invention is applied to a color image, the luminance signal and the chrominance signal may be separately processed, or the luminance signal may be processed first, and the chrominance signal may be used as a result of the distortion removal characteristic of the luminance signal. The processing may be determined using the information. Further, the same processing may be performed on both signals by selecting the preferred one of the distortion removal characteristics determined for both signals.

Further, the kernel size of the filter is arbitrary, and data for determining the filter characteristics prior to the filtering and all the reproduced data of each block are obtained and stored in a memory for filtering. You may do it.

Further, the present invention is not limited to the block size, the type of orthogonal transform, the type of variable length coding, and the like used in the above-described embodiments. The filter may be applied not to the entire block but only to the vicinity of the block boundary, or any filter other than the filter may be used as long as it has a distortion removing effect.

[0029]

According to the present invention having the above-described structure, the block distortion can be adaptively removed using only the transmitted or recorded image information, and the circuit is simple in structure. Therefore, the present invention can be applied not only to still images but also to digital electronic cameras with a function of reproducing moving images, etc.

Further, according to the present invention, the encoding device may have the conventional configuration. That is, the effect can be improved only by devising the decoding device for the standard compression method, and the reproduction can be performed as usual, and the degree of distortion removal can be freely set.

[Brief description of the drawings]

FIG. 1 is a circuit configuration diagram of a first embodiment of an image signal decoding device according to the present invention.

FIGS. 2A and 2B are diagrams showing regions where significant coefficients of blocks exist.

3A is a diagram for explaining blocking, and FIG. 3B is a diagram for explaining DCT coefficients.

FIG. 4 is a circuit configuration diagram of a third embodiment of the image signal decoding device of the present invention.

FIGS. 5A and 5B are diagrams for explaining weights for each coefficient.

FIG. 6 is a diagram illustrating characteristics of an ideal filter.

FIG. 7 is a diagram showing a configuration of a conventional encoding / decoding device.

[Explanation of symbols]

11: variable length code decoding circuit, 12: inverse quantization circuit,
13: an inverse orthogonal transform circuit, 14: a distortion removal circuit, 15: a code amount counting circuit in a block, 16: a distortion removal characteristic determination circuit.

Claims (1)

(57) [Claims] An image data is divided into blocks, an orthogonal transform is performed for each of the divided blocks, and the transformed output is quantized by a quantizing means. Thereafter, the quantized output is supplied to a variable length coding means. Variable-length code decoding means for decoding image data compressed by variable-length coding and compression, inverse quantization means for inversely quantizing the decoded output from the variable-length code decoding means, and Inverse orthogonal transform means for performing inverse orthogonal transform of the inverse quantized output of the above, distortion removing means for performing a distortion removing process on the transform output from the inverse orthogonal transform means, and the number of significant coefficients of transform coefficients of each block. Or the distortion removal characteristic of the distortion removing means is changed based on any of the sums of weights determined in advance by respective sequences of significant coefficients of the transform coefficients of each block. Image signal decoding apparatus characterized by comprising a strain relief characterization means.