CN115035556A - Face retrieval method and device, electronic equipment and storage medium - Google Patents

Abstract

The invention provides a face retrieval method, device, electronic equipment and storage medium. The method includes: performing feature extraction on a face image to be retrieved, and obtaining a feature vector of the face image to be retrieved as the feature vector to be retrieved; The similarity between the feature vector to be retrieved and the feature vector of the bottom library picture in the base library; according to the distribution information between the largest top N similarities in each similarity and/or the maximum similarity in each similarity , the original similarity threshold is adjusted to obtain the adjusted similarity threshold; the retrieval result is determined according to the maximum similarity and the adjusted similarity threshold. The present invention fully considers the differences caused by different image qualities, so that the difference in retrieval precision under different image qualities can be reduced, the retrieval accuracy under different image qualities can be improved, and the recall rate of face retrieval can be improved.

Description

Face retrieval method, device, electronic device and storage medium

technical field

The present invention relates to the technical field of face recognition, in particular to a face retrieval method, device, electronic device and storage medium.

Background technique

The face recognition method is to encode the face pictures through a deep neural network model, and measure whether the two pictures are the same person according to the similarity between the encoded features. In the face retrieval scenario, the algorithm needs to give whether the face in the face image to be retrieved exists in the base, and which image in the base. The current algorithm is calculated in two steps: 1) Calculate the similarity between the feature corresponding to the face image to be retrieved and the corresponding feature of each face image in the base library, and find the feature in the base library that is similar to the face image to be retrieved 2) Determine whether the maximum similarity is greater than a certain threshold, if it is greater than the threshold, it is considered that the face of the face image to be retrieved exists in the base library, and the corresponding base library face image is output, If it is less than the threshold, it is considered that the face of the face image to be retrieved does not exist in the base library, and no matching image is output.

In the prior art, a fixed similarity threshold is used to determine whether two face images are of the same person, which cannot solve the shift of the similarity distribution caused by the characteristics of the face images to be retrieved. For example, a clear face picture to be retrieved can be searched for its pictures in the base library according to the prior art method, but if the face picture to be searched becomes blurred, the similarity with all the pictures in the base library will become smaller. Although the image of the same person is still matched with the highest similarity in the base library, the overall similarity cannot exceed the threshold due to the smaller offset of the overall similarity, so the correct base image cannot be output, which makes the algorithm as a whole The recall rate becomes lower.

SUMMARY OF THE INVENTION

In view of the above problems, the embodiments of the present invention are proposed to provide a face retrieval method, apparatus, electronic device and storage medium that overcome the above problems or at least partially solve the above problems.

According to a first aspect of the embodiments of the present invention, a face retrieval method is provided, including:

Perform feature extraction on the face image to be retrieved, and obtain the feature vector of the face image to be retrieved as the feature vector to be retrieved;

respectively determining the similarity between the feature vector to be retrieved and the feature vector of the bottom library picture in the bottom library;

Adjust the original similarity threshold according to the distribution information between the largest top N similarities in each similarity and/or the maximum similarity in each similarity to obtain the adjusted similarity threshold;

The retrieval result is determined according to the maximum similarity and the adjusted similarity threshold.

According to a second aspect of the embodiments of the present invention, a face retrieval apparatus is provided, including:

The feature extraction module is used to perform feature extraction on the face image to be retrieved, and obtain the feature vector of the face image to be retrieved as the feature vector to be retrieved;

a similarity determination module, used for respectively determining the similarity between the feature vector to be retrieved and the feature vector of the bottom library picture in the bottom library;

The threshold adjustment module is used to adjust the original similarity threshold according to the distribution information between the largest top N similarities in each similarity and/or the maximum similarity in each similarity, and obtain the adjusted similarity threshold ;

The retrieval result determination module is configured to determine the retrieval result according to the maximum similarity and the adjusted similarity threshold.

According to a third aspect of the embodiments of the present invention, there is provided an electronic device, comprising: a processor, a memory, and a computer program stored on the memory and executable on the processor, the computer program being When executed by the processor, the face retrieval method as described in the first aspect is implemented.

According to a fourth aspect of the embodiments of the present invention, there is provided a computer-readable storage medium, where a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the first aspect is implemented face retrieval method.

In the face retrieval method, device, electronic device, and storage medium provided by the embodiments of the present invention, the feature extraction of the face image to be retrieved is performed to obtain the feature vector of the face image to be retrieved, and the feature vector to be retrieved is used as the feature vector to be retrieved, and the feature to be retrieved is determined respectively. The similarity between the vector and the feature vector of the bottom library image in the base library, according to the distribution information between the largest top N similarities in each similarity and/or the maximum similarity in each similarity, the original similarity The threshold is adjusted to obtain the adjusted similarity threshold, and the retrieval result is determined according to the maximum similarity and the adjusted similarity threshold. Since the distribution information between the largest top N similarities and/or the largest similarity in each similarity is determined. Similarity is used to adjust the original similarity threshold, fully considering the differences caused by different image qualities, so that the difference in retrieval accuracy under different image qualities can be reduced, and the retrieval accuracy under different image qualities can be improved. retrieval recall.

The above description is only an overview of the technical solutions of the present invention, in order to be able to understand the technical means of the present invention more clearly, it can be implemented according to the content of the description, and in order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand , the following specific embodiments of the present invention are given.

Description of drawings

Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The drawings are for the purpose of illustrating preferred embodiments only and are not to be considered limiting of the invention.

1 is a flowchart of steps of a method for retrieving a face provided by an embodiment of the present invention;

FIG. 2 is a structural block diagram of a face retrieval and detection apparatus provided by an embodiment of the present invention.

Detailed ways

Exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention may be embodied in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided so that the present invention will be more thoroughly understood, and will fully convey the scope of the present invention to those skilled in the art.

FIG. 1 is a flowchart of steps of a face retrieval method provided by an embodiment of the present invention. As shown in FIG. 1 , the method may include:

Step 101: Perform feature extraction on the to-be-retrieved face picture to obtain a feature vector of the to-be-retrieved face picture as the to-be-retrieved feature vector.

The feature extraction is performed on the face image to be retrieved through the neural network model, the feature vector of the face image to be retrieved is obtained, and the feature vector of the face image to be retrieved is used as the feature vector to be retrieved.

Step 102, respectively determining the similarity between the feature vector to be retrieved and the feature vector of the bottom library image in the bottom library.

Through the neural network model, the features of the face images to be retrieved are extracted respectively, and the feature vectors of the images in the base library and the base library are obtained. The similarity between the feature vector to be retrieved and the feature vector of the bottom library image in the bottom library is determined respectively. The similarity may be a cosine similarity, or may also be a similarity expressed by a distance. The similarity is inversely proportional to the distance. The smaller the distance between the two vectors, the greater the similarity. The greater the distance, the greater the similarity. is small, so that the largest similarity corresponds to the smallest distance. The distance may be the L2 distance or the like.

The feature extraction of each base image in the base library can be carried out in advance through the neural network model to obtain the feature vector of the base image in the base library, and the feature vector of each base image is saved, so as to obtain the to-be-retrieved feature vector. When using a face image, the similarity between the to-be-retrieved feature vector corresponding to the to-be-retrieved face image and the stored feature vector of each base image can be directly calculated, which improves the face retrieval speed and saves the face retrieval time.

Step 103: Adjust the original similarity threshold according to the distribution information between the largest top N similarities and/or the largest similarity in each similarity to obtain an adjusted similarity threshold.

After determining the similarity between the feature vector to be retrieved and the feature vector of the base image in the base library, compare these similarities, determine the largest top N similarities, and determine the distribution information of the top N similarities, such as The maximum similarity is much larger than the 2-Nth similarity, or the maximum similarity is slightly larger than the 2-Nth similarity, or the maximum similarity is relatively close to the 2-Nth similarity, etc. The distribution information between the N similarities and/or the maximum similarity among the similarities is adjusted to obtain the adjusted original similarity threshold. Among them, N is a preset number, for example, it can be 10. In one example, a judgment standard that is much larger, slightly larger, or relatively close can be preset. For example, if the maximum similarity is more than 10% greater than the second similarity, it can be considered that the maximum similarity is far greater than the 2-Nth similarity; the difference between the maximum similarity and the second similarity is less than 10% and greater than 3% , it can be considered that the maximum similarity is slightly larger than the 2-Nth similarity, otherwise it is considered that the maximum similarity is relatively close to the 2-Nth similarity. It can be considered that the maximum similarity is slightly larger than the 2-Nth similarity, or when the maximum similarity is relatively close to the 2-Nth similarity, the maximum similarity is not much larger than the 2-Nth similarity. It can also be considered that when the maximum similarity is relatively close to the 2-Nth similarity, the maximum similarity is not much larger than the 2-Nth similarity.

In an embodiment of the present invention, determining the largest top N similarities from the similarities, optionally including: sorting the similarities in descending order of the similarities; Among the similarities, the first N similarities are determined, and the first similarity is determined as the maximum similarity.

After determining the similarity between the feature vector to be retrieved and the feature vector of the base image in the base library, sort all the obtained similarities in descending order, and select N similarities from the first position after sorting. The similarity ranked first is the maximum similarity. Wherein, the N is greater than or equal to 2.

In another embodiment of the present invention, determining the largest top N similarities from among the similarities may optionally include: sorting the similarities in ascending order of the similarities; Among the similarities, the following N similarities are determined, and the last similarity is determined as the maximum similarity.

After determining the similarity between the feature vector to be retrieved and the feature vector of the base image in the base library, sort all the obtained similarities in ascending order. The similarity in the last digit is the maximum similarity. Wherein, the N is greater than or equal to 2.

In an embodiment of the present invention, the original similarity threshold is adjusted according to the distribution information between the largest top N similarities in each similarity and/or the largest similarity in each similarity to obtain the adjusted similarity threshold. Similarity thresholds, including:

If the maximum similarity is far greater than the 2nd-Nth similarity, the original similarity threshold is adjusted so that the adjusted similarity threshold is smaller than the original similarity threshold.

Among them, far greater than means that the distribution information is greater than the distribution threshold.

If the maximum similarity is far greater than the 2-Nth similarity, it means that the base image corresponding to the maximum similarity is likely to be the base image corresponding to the image to be retrieved. Decrease the original similarity threshold.

For example, if the original similarity threshold is 70% and the maximum similarity is 70.2%, the similarity in the second place is 48% and the similarity in the third place is 47% after sorting in descending order. , so as to determine that the distribution information of the similarity is that the maximum similarity is far greater than the similarity between the to-be-retrieved picture and other base images, the original similarity threshold can be appropriately reduced to determine the base image corresponding to the maximum similarity in the ratio of the images to be searched. .

In another embodiment of the present invention, the original similarity threshold is adjusted according to the distribution information between the largest top N similarities in each similarity and/or the largest similarity in each similarity, and the adjusted threshold is obtained after adjustment. The similarity threshold of , including:

If the maximum similarity is far greater than the 2nd-Nth similarity and the maximum similarity is smaller than the original similarity threshold, adjust the original similarity threshold so that the adjusted similarity threshold is smaller than the original similarity threshold.

If the maximum similarity is far greater than the 2nd-Nth similarity, and the maximum similarity is smaller than the original similarity threshold, it means that the maximum similarity caused by the image quality cannot exceed the original similarity threshold. At this time, in order to avoid this The original similarity threshold can be adjusted so that the adjusted similarity threshold is smaller than the original similarity threshold.

For example, if the original similarity threshold is 70%, the maximum similarity is 69%, the similarity in the second place after sorting from largest to smallest is 40%, and the similarity in the third place is 38% , so that the distribution information of the similarity is determined that the maximum similarity is far greater than the similarity between the to-be-retrieved picture and other base images, which means that the possibility of the base image corresponding to the maximum similarity of the to-be-retrieved image is increased. Appropriately lower the original similarity threshold.

In an embodiment of the present invention, the original similarity threshold is adjusted according to the distribution information between the largest top N similarities in each similarity and/or the largest similarity in each similarity to obtain the adjusted similarity threshold. Similarity thresholds, including:

If the maximum similarity is not far greater than the 2nd-Nth similarity, the original similarity threshold is adjusted so that the adjusted similarity threshold is greater than the original similarity threshold.

If the maximum similarity is not much larger than the 2nd-Nth similarity, that is, the distribution information is smaller than the distribution threshold, that is, each similarity is relatively close, the original similarity threshold can be appropriately increased, even if the adjusted similarity threshold is obtained greater than the original similarity threshold.

In another embodiment of the present invention, the original similarity threshold is adjusted according to the distribution information between the largest top N similarities in each similarity and/or the largest similarity in each similarity, and the adjusted threshold is obtained after adjustment. The similarity threshold of , including:

If the maximum similarity is not far greater than the 2nd-Nth similarity and the maximum similarity is greater than the original similarity threshold, adjust the original similarity threshold so that the adjusted similarity threshold is greater than the original similarity degree threshold.

If the maximum similarity is not much greater than the 2nd-Nth similarity, that is, the distribution information is less than the distribution threshold, that is, each similarity is relatively close, then if the maximum similarity is greater than the original similarity threshold, there may be a recall error In order to avoid this problem, the original similarity threshold can be appropriately increased, that is, the adjusted similarity threshold is larger than the original similarity threshold.

For example, if the original similarity threshold is 70%, the maximum similarity is 70.2%, the similarity in the second place after sorting from largest to smallest is 68%, and the similarity in the third place is 67% , so as to determine that the similarity distribution information is that each similarity is relatively close, then although the maximum similarity is higher than the original similarity threshold, there is also a certain probability of misrecognition (for example, the person corresponding to the image to be retrieved is not in the base library). The original similarity threshold can be appropriately increased.

In an embodiment of the present invention, the distribution information is determined by the following steps: determining a reference similarity according to the largest top N similarities in each similarity; determining a reference similarity according to the maximum similarity and the reference similarity the distribution information.

When the N is 2, the second similarity may be used as the reference similarity, and when N>2, the reference similarity may be determined according to the maximum similarity and the 2-Nth similarity respectively. After the reference similarity is determined, distribution information can be obtained based on the maximum similarity and the reference similarity. Afterwards, the original similarity threshold may be adjusted in a preset manner according to the respective information to obtain an adjusted similarity threshold. By determining the reference similarity and distribution information, and adjusting the original similarity threshold based on the distribution information, it is possible to make full use of the distribution change of the feature similarity caused by the characteristics of the image to be retrieved, so that the original similarity threshold can be adjusted more appropriately. To improve the recall rate of face retrieval.

In an optional implementation manner, the determining the reference similarity according to the largest top N similarities in each similarity includes:

determining the second similarity as the reference similarity; or

Taking the 2-Nth similarity as the reference similarity; or

The average value of the 2-Nth similarity is taken as the reference similarity.

The second similarity can be used as the reference similarity, or the 2-N similarity can be used as the reference similarity, or the average value of the 2-N similarity can be used as the reference similarity. Clearly reflects the distribution of the 2-Nth similarity.

In an optional implementation manner, determining the distribution information according to the maximum similarity and the reference similarity includes: determining the distribution information according to a difference between the maximum similarity and the reference similarity distribution information; or, determining the distribution information according to the ratio of the maximum similarity to the reference similarity.

Taking the difference between the maximum similarity and the reference similarity as distribution information, the value of the maximum similarity that is more than at least one similarity can be clearly calculated, so that the original similarity threshold can be adjusted according to this value. For example, the larger the distribution information, the lower the original similarity threshold; the smaller the distribution information, the closer the maximum similarity and other similarities are, in this case, the original similarity threshold can be appropriately increased.

In an example, the distribution information is determined according to the difference between the maximum similarity and the reference similarity, and the relationship between the distribution information and the maximum similarity and the reference similarity is expressed as follows:

distgap=s[0]-f(s)

Among them, distgap is the distribution information, s[0] is the maximum similarity, f(s) is the reference similarity, and the reference similarity can be expressed in one of the following two ways:

f(s)=s[1]

Among them, f(s) is the reference similarity function, s[1] is the second similarity, s[i] is the i+1th similarity, and n is N-1.

In addition to the above methods that can represent the distribution information of the similarity, the ratio of the maximum similarity to the reference similarity can also be determined as the distribution information, and the multiple of the maximum similarity relative to the 2-Nth similarity can be clearly calculated. The original similarity threshold is adjusted according to this value. For example, the larger the similarity distribution ratio, the lower the original similarity threshold; the smaller the similarity distribution ratio, which means that the maximum similarity and the 2nd-Nth similarity are relatively close, and the original similarity threshold can be appropriately increased. .

In an optional implementation manner, the original similarity threshold is adjusted according to the distribution information between the largest top N similarities in each similarity and/or the largest similarity in each similarity, and the adjusted threshold is obtained. The similarity threshold value includes: adjusting the original similarity threshold according to the product of the distribution information and the distribution adjustment coefficient and/or the product of the maximum similarity and the maximum similarity adjustment coefficient, and obtaining the adjusted similarity threshold. similarity threshold.

Wherein, the distribution adjustment coefficient and/or the maximum similarity adjustment coefficient are obtained through verification on a calibration data set.

The face retrieval method can be verified on the calibration data set in advance, and the distribution adjustment coefficient or the maximum similarity adjustment coefficient with the best performance on the recall rate, misrecognition rate and other indicators can be fitted. After the distribution adjustment coefficient is obtained, it can be applied to the adjustment of the original similarity threshold. The original similarity threshold can be adjusted according to the following formula:

th _new =th+k*distgap

Wherein, th _new is the adjusted similarity threshold, th is the original similarity threshold, k is a distribution adjustment coefficient, and distgap is distribution information. That is, when adjusting the original similarity threshold, the product of the distribution information and the distribution adjustment coefficient is used as the adjustment value of the original similarity threshold, and then the original similarity threshold and the adjustment value are added to obtain the adjusted similarity threshold.

When adjusting the original similarity threshold, the product of the maximum similarity and the maximum similarity adjustment coefficient can also be used as the adjustment value of the original similarity threshold, and then the original similarity threshold and the adjustment value are added to obtain the adjusted similarity threshold.

When adjusting the original similarity threshold, the product of the distribution information and the distribution adjustment coefficient can also be used as the first adjustment value of the original similarity threshold, and the product of the maximum similarity and the maximum similarity adjustment coefficient can be used as the original similarity threshold. The second adjustment value, and then adding the original similarity threshold, the first adjustment value and the second adjustment value to obtain the adjusted similarity threshold.

Through the distribution adjustment coefficient and the maximum similarity adjustment coefficient, the dynamic adjustment of the original similarity threshold is realized.

Step 104: Determine a retrieval result according to the maximum similarity and the adjusted similarity threshold.

The maximum similarity is compared with the adjusted original similarity threshold, and the retrieval result is determined according to the comparison result.

In an embodiment of the present invention, determining the retrieval result according to the maximum similarity and the adjusted similarity threshold includes: if the maximum similarity is greater than or equal to the adjusted similarity threshold, then The base image corresponding to the maximum similarity is used as the retrieval result.

After adjusting the original similarity threshold, if the maximum similarity is greater than or equal to the adjusted similarity threshold, the base image corresponding to the maximum similarity is used as the retrieval result. If the maximum similarity is less than the adjusted similarity threshold, Then it is determined that the search result is that there is no matching picture in the base library.

In the face retrieval method provided in this embodiment, a feature vector of the face image to be retrieved is obtained by extracting the feature of the face image to be retrieved, which is used as the feature vector to be retrieved, and the feature vector to be retrieved and the image in the base library are determined respectively. For the similarity between feature vectors, adjust the original similarity threshold according to the distribution information between the largest top N similarities in each similarity and/or the largest similarity in each similarity, and obtain the adjusted similarity The degree threshold, the retrieval result is determined according to the maximum similarity and the adjusted similarity threshold, since the original similarity threshold is determined according to the distribution information between the largest top N similarities and/or the maximum similarity in each similarity. The adjustment fully considers the differences caused by different image qualities, so that the difference in retrieval accuracy under different image qualities can be reduced, the retrieval accuracy under different image qualities can be improved, and the recall rate of face retrieval can be improved.

It should be noted that, for the sake of simple description, the method embodiments are described as a series of action combinations, but those skilled in the art should know that the embodiments of the present invention are not limited by the described action sequences, because According to embodiments of the present invention, certain steps may be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily required by the embodiments of the present invention.

FIG. 2 is a structural block diagram of a face retrieval apparatus provided by an embodiment of the present invention. As shown in FIG. 2 , the face retrieval apparatus may include:

The feature extraction module 201 is configured to perform feature extraction on the to-be-retrieved face picture to obtain a feature vector of the to-be-retrieved face picture as the to-be-retrieved feature vector;

A similarity determination module 202, configured to respectively determine the similarity between the feature vector to be retrieved and the feature vector of the bottom library image in the bottom library;

Threshold adjustment module 203, configured to adjust the original similarity threshold according to the distribution information between the largest top N similarities in each similarity and/or the maximum similarity in each similarity to obtain the adjusted similarity threshold;

The retrieval result determination module 204 is configured to determine the retrieval result according to the maximum similarity and the adjusted similarity threshold.

Optionally, the threshold adjustment module includes:

The first adjustment unit is configured to adjust the original similarity threshold if the maximum similarity is far greater than the 2-Nth similarity, so that the adjusted similarity threshold is smaller than the original similarity threshold.

Optionally, the threshold adjustment module includes:

The second adjustment unit is configured to adjust the original similarity threshold if the maximum similarity is far greater than the 2-Nth similarity and the maximum similarity is smaller than the original similarity threshold, so that the adjusted The similarity threshold is smaller than the original similarity threshold.

Optionally, the threshold adjustment module includes:

The third adjustment unit, if the maximum similarity is not far greater than the 2nd-Nth similarity, adjusts the original similarity threshold so that the adjusted similarity threshold is greater than the original similarity threshold.

Optionally, the threshold adjustment module includes:

The fourth adjustment unit, if the maximum similarity is not far greater than the 2nd-Nth similarity and the maximum similarity is greater than the original similarity threshold, adjust the original similarity threshold to make the adjusted similarity The degree threshold is greater than the original similarity threshold.

Optionally, the device further includes:

The reference similarity determination module is used to determine the reference similarity according to the largest top N similarities in each similarity;

A distribution information determination module, configured to determine the distribution information according to the maximum similarity and the reference similarity.

Optionally, the reference similarity determination module is specifically used for:

determining the second similarity as the reference similarity; or

Taking the 2-Nth similarity as the reference similarity; or,

The average value of the 2-Nth similarity is taken as the reference similarity.

Optionally, the distribution information determination module is specifically used for:

Determine the distribution information according to the difference between the maximum similarity and the reference similarity; or,

The distribution information is determined according to the ratio of the maximum similarity to the reference similarity.

Optionally, the threshold adjustment module is specifically used for:

The original similarity threshold is adjusted according to the product of the distribution information and the distribution adjustment coefficient and/or the product of the maximum similarity and the maximum similarity adjustment coefficient to obtain an adjusted similarity threshold.

Wherein, the distribution adjustment coefficient and/or the maximum similarity adjustment coefficient are obtained through verification on a calibration data set.

Optionally, the retrieval result determination module is specifically used for:

If the maximum similarity is greater than or equal to the adjusted similarity threshold, the base image corresponding to the maximum similarity is used as the retrieval result.

In the face retrieval apparatus provided in this embodiment, the feature extraction module performs feature extraction on the face image to be retrieved, and the feature vector of the face image to be retrieved is obtained as the feature vector to be retrieved, and the similarity determination module determines the feature vector to be retrieved and The similarity between the feature vectors of the bottom library images in the base library, the threshold adjustment module, according to the distribution information between the largest top N similarities in each similarity and/or the maximum similarity in each similarity, to the original similarity Adjust the degree threshold to obtain the adjusted similarity threshold, and the retrieval result determination module determines the retrieval result according to the maximum similarity and the adjusted similarity threshold. The maximum similarity in the similarity is used to adjust the original similarity threshold, and the difference caused by different image quality is fully considered, so that the difference in retrieval accuracy under different image quality can be reduced and the retrieval accuracy under different image quality can be improved. Thus, the recall rate of face retrieval can be improved.

As for the apparatus embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and reference may be made to the partial description of the method embodiment for related parts.

Further, according to an embodiment of the present invention, an electronic device is provided, the electronic device may be a terminal device such as a computer, or a server, etc., the electronic device includes: a processor, a memory, and a A computer program on the memory and executable on the processor, when the computer program is executed by the processor, implements the face retrieval method of the foregoing embodiment.

According to an embodiment of the present invention, a computer-readable storage medium is also provided, the computer-readable storage medium includes but not limited to magnetic disk storage, CD-ROM, optical storage, etc., on which storage medium is stored There is a computer program that, when executed by a processor, implements the face retrieval method of the foregoing embodiment.

The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same and similar parts between the various embodiments may be referred to each other.

It should be understood by those skilled in the art that the embodiments of the embodiments of the present invention may be provided as a method, an apparatus, or a computer program product. Accordingly, embodiments of the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, embodiments of the present invention may take the form of a computer program product implemented on one or more computer-usable storage media having computer-usable program code embodied therein, including but not limited to disk storage, CD-ROM, optical storage, and the like.

Embodiments of the present invention are described with reference to flowcharts and/or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the present invention. It will be understood that each flow and/or block in the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to the processor of a general purpose computer, special purpose computer, embedded processor or other programmable data processing terminal equipment to produce a machine that causes the instructions to be executed by the processor of the computer or other programmable data processing terminal equipment Means are created for implementing the functions specified in the flow or flows of the flowcharts and/or the blocks or blocks of the block diagrams.

These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing terminal equipment to operate in a particular manner, such that the instructions stored in the computer readable memory result in an article of manufacture comprising instruction means, the The instruction means implement the functions specified in the flow or flow of the flowcharts and/or the block or blocks of the block diagrams.

These computer program instructions can also be loaded on a computer or other programmable data processing terminal equipment, so that a series of operational steps are performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby executing on the computer or other programmable terminal equipment The instructions executed on the above provide steps for implementing the functions specified in the flowchart or blocks and/or the block or blocks of the block diagrams.

Although preferred embodiments of the embodiments of the present invention have been described, additional changes and modifications to these embodiments may be made by those skilled in the art once the basic inventive concepts are known. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications that fall within the scope of the embodiments of the present invention.

Finally, it should also be noted that in this document, relational terms such as first and second are used only to distinguish one entity or operation from another, and do not necessarily require or imply these entities or that there is any such actual relationship or sequence between operations. Moreover, the terms "comprising", "comprising" or any other variation thereof are intended to encompass non-exclusive inclusion, such that a process, method, article or terminal device comprising a list of elements includes not only those elements, but also a non-exclusive list of elements. other elements, or also include elements inherent to such a process, method, article or terminal equipment. Without further limitation, an element defined by the phrase "comprises a..." does not preclude the presence of additional identical elements in the process, method, article or terminal device comprising said element.

The above provides a detailed introduction to a face retrieval method, device, electronic device and storage medium provided by the present invention. In this paper, specific examples are used to illustrate the principles and implementations of the present invention. The descriptions of the above embodiments are only It is used to help understand the method of the present invention and its core idea; at the same time, for those skilled in the art, according to the idea of the present invention, there will be changes in the specific embodiments and application scope. The contents of the description should not be construed as limiting the present invention.

Claims (14)

1. A face retrieval method is characterized by comprising the following steps:

extracting features of the face picture to be retrieved to obtain a feature vector of the face picture to be retrieved, and using the feature vector as a feature vector to be retrieved;

respectively determining the similarity between the feature vector to be retrieved and the feature vector of the base picture in the base library;

adjusting the original similarity threshold according to the distribution information among the first N largest similarities in all the similarities and/or the largest similarity in all the similarities to obtain an adjusted similarity threshold;

and determining a retrieval result according to the maximum similarity and the adjusted similarity threshold.

2. The method of claim 1, wherein adjusting the original similarity threshold according to distribution information between the top N largest similarities in the similarities and/or the largest similarities in the similarities to obtain an adjusted similarity threshold comprises:

and if the maximum similarity is far larger than the 2 nd-Nth similarity, adjusting the original similarity threshold value to ensure that the adjusted similarity threshold value is smaller than the original similarity threshold value.

3. The method of claim 1, wherein adjusting the original similarity threshold according to distribution information between the top N largest similarities in the similarities and/or the largest similarities in the similarities to obtain an adjusted similarity threshold comprises:

and if the maximum similarity is far larger than the 2 nd-Nth similarity and the maximum similarity is smaller than the original similarity threshold, adjusting the original similarity threshold to ensure that the adjusted similarity threshold is smaller than the original similarity threshold.

4. The method according to any one of claims 1 to 3, wherein the adjusting the original similarity threshold according to the distribution information between the top N largest similarities in the similarities and/or the largest similarities in the similarities to obtain an adjusted similarity threshold comprises:

and if the maximum similarity is not far greater than the 2 nd-Nth similarity, adjusting the original similarity threshold value to ensure that the adjusted similarity threshold value is greater than the original similarity threshold value.

5. The method according to any one of claims 1 to 3, wherein the adjusting the original similarity threshold according to the distribution information between the top N largest similarities in the similarities and/or the largest similarities in the similarities to obtain an adjusted similarity threshold comprises:

if the maximum similarity is not far greater than the 2 nd-Nth similarity and the maximum similarity is greater than the original similarity threshold, adjusting the original similarity threshold so that the adjusted similarity threshold is greater than the original similarity threshold.

6. The method according to any of claims 1-5, wherein the distribution information is determined by:

determining reference similarity according to the maximum first N similarity in the similarities;

and determining the distribution information according to the maximum similarity and the reference similarity.

7. The method of claim 6, wherein determining the reference similarity according to the top N largest similarities among the similarities comprises:

determining the 2 nd similarity as the reference similarity; or

Taking the 2-N similarity as the reference similarity; or,

the average of the 2 nd-N number of similarities is taken as the reference similarity.

8. The method according to claim 6 or 7, wherein determining the distribution information according to the maximum similarity and the reference similarity comprises:

determining the distribution information according to the difference value between the maximum similarity and the reference similarity; or,

and determining the distribution information according to the ratio of the maximum similarity to the reference similarity.

9. The method according to any one of claims 1 to 8, wherein adjusting the original similarity threshold according to the distribution information between the top N largest similarities in the similarities and/or the largest similarities in the similarities to obtain an adjusted similarity threshold comprises:

and adjusting the original similarity threshold according to the product of the distribution information and a distribution adjustment coefficient and/or the product of the maximum similarity and the maximum similarity adjustment coefficient to obtain an adjusted similarity threshold.

10. The method according to claim 9, wherein the distribution adjustment factor and/or the maximum similarity adjustment factor is obtained by verification on a correction data set.

11. The method according to any one of claims 1-10, wherein determining a search result according to the maximum similarity and the adjusted similarity threshold comprises:

and if the maximum similarity is larger than or equal to the adjusted similarity threshold, taking the bottom library picture corresponding to the maximum similarity as the retrieval result.

12. A face retrieval apparatus, comprising:

the characteristic extraction module is used for extracting the characteristics of the face picture to be retrieved to obtain the characteristic vector of the face picture to be retrieved as the characteristic vector to be retrieved;

the similarity determining module is used for respectively determining the similarity between the feature vector to be retrieved and the feature vector of the base image in the base;

the threshold adjusting module is used for adjusting the original similarity threshold according to the distribution information among the first N maximum similarities in all the similarities and/or the maximum similarity in all the similarities to obtain an adjusted similarity threshold;

and the retrieval result determining module is used for determining the retrieval result according to the maximum similarity and the adjusted similarity threshold.

13. An electronic device, comprising: a processor, a memory and a computer program stored on the memory and executable on the processor, the computer program, when executed by the processor, implementing the face retrieval method of any one of claims 1-11.

14. A computer-readable storage medium, characterized in that a computer program is stored thereon, which computer program, when being executed by a processor, carries out a face retrieval method according to any one of claims 1-11.

Images