Computer portrait recognition system

Abstract

The invention discloses a computer face recognition system, and relates to the technical field of face recognition systems. The computer portrait recognition system comprises: an image acquisition unit; acquiring a face photo based on optical sensing equipment; an eyeball basic recognition template unit; screening pictures which cannot be preliminarily recognized in the input sample picture set according to the eyeball recognition result; an identification unit; and carrying out average value calculation on the feature vectors of the image set, and introducing the obtained average value feature vector into a neighbor classifier to be compared with the feature vector of the established face library. According to the invention, the obtained mean value feature vector of the screened picture set is introduced into the neighbor classifier and is compared with the feature vector of the built face library, so that the portrait recognition of the picture set type is realized, and on the premise of realizing directional recognition, the rapid recognition of a large number of images to which a single face belongs is completed, and the face recognition efficiency is improved. Therefore, the system can cope with random identification tests and can realize identification tasks in most scenes.

Description

Technical Field

[0001] The present invention relates to the technical field of face recognition systems, and specifically to a computer portrait recognition system. Background Art

[0002] Face recognition is a biometric identification technology based on the facial feature information of a person. A series of related technologies, such as using a camera or webcam to collect images or video streams containing human faces, automatically detecting and tracking human faces in the images, and then performing facial recognition on the detected human faces, are usually also called portrait recognition or facial recognition.

[0003] Broadly speaking, portrait recognition actually includes a series of related technologies for building a portrait recognition system, including face image acquisition, face localization, portrait recognition preprocessing, identity confirmation, and identity search, etc.; while narrowly speaking, portrait recognition specifically refers to the technology or system for confirming or searching for identities through human faces. Although portrait recognition has many advantages that cannot be compared with other identifications, it also has many difficulties itself. Portrait recognition is considered to be one of the most difficult research topics in the field of biometric identification and even in the field of artificial intelligence. The difficulties of portrait recognition are mainly brought about by the characteristics of the human face as a biometric feature.

[0004] For example, as disclosed in a computer portrait recognition system with the Chinese invention publication number CN111680280B, the recognition system splits the data according to the face image collected by the image acquisition unit, then performs virtual modeling, and outputs the facial feature parameters on the virtual model; then performs a directional search in the face image data according to the feature parameters of the portrait modeling unit, analyzes the correlation between its virtual model and the search result through the model analysis unit, combines with the eigenvalue comparison module for comparison, and obtains the initial screening result; according to the target image in the initial screening result, performs a differential analysis, analyzes the secondary feature difference between the screened image and the original data image, performs a screening search again in the data management unit according to the secondary eigenvalue, and performs a comparison according to the secondary screening object to obtain the final refined result.

[0005] The above recognition system performs prior modeling on the face images collected by the image acquisition unit before face recognition, so as to facilitate subsequent targeted searches. Similar to many existing systems, the above recognition system docks and recognizes individual photos collected by the data acquisition module. Such a recognition method is suitable for targeted recognition by research institutions and important units. However, under random shooting during image acquisition by the image acquisition unit, a large number of images of single faces often appear. When using the above method to recognize these images, the acquisition end needs to perform recognition of multiple groups of images, flipping and cropping of the best images, and interception, and then perform subsequent recognition steps, resulting in a reduction in the recognition efficiency of the overall recognition system. It is not suitable for rapid portrait recognition in public areas such as streets, public places, and surveillance sites, and cannot handle the randomness of the pictures collected by the image acquisition end. Summary of the Invention

[0006] Aiming at the deficiencies of the prior art, the present invention provides a computer portrait recognition system, which solves the problems put forward in the above background technology.

[0007] To achieve the above objectives, the present invention is realized through the following technical solutions. A computer portrait recognition system, the system includes:

[0008] An image acquisition unit; collecting face photos based on an optical sensing device, using the face photos as the input of the computer, and completing the basic processing of the input image based on digital image processing to obtain the required sample picture set;

[0009] An eyeball basic recognition template unit; based on the obtained sample picture set, selecting the face area as the face sample, performing Sobel edge extraction on all the extracted face samples, averaging the gray levels of the obtained edge images and transforming them to the required size as the original eyeball recognition template, screening the pictures in the input sample picture set that cannot be initially recognized according to the eyeball recognition result, and feeding the recognition failure result back to the image acquisition unit through the computer;

[0010] A recognition unit; the recognition unit completes the LBP feature extraction of the sample picture set based on the feature extraction sub-unit, obtains the image set feature vector of the sample picture set, calculates the average value of the image set feature vector, and introduces the obtained average value feature vector into the nearest neighbor classifier to compare with the feature vector of the established face database.

[0011] A further improvement of the technical solution of the present invention lies in that the basic processing of the input image based on digital image processing includes: quantifying and sampling the spatial coordinates of the continuous input image function based on the computer to obtain the digital image set of the input image set;

[0012] The computer composes the pixels in the obtained digital image in different ways, and performs noise filtering, image sharpening, contrast enhancement, and edge detection on the digital image to obtain the required sample picture set;

[0013] Among them, for optical sensing devices such as USB cameras, the Linux system and V4L2 are adopted in the computer system to provide an image acquisition interface. After the image acquisition unit performs noise filtering, image sharpening, contrast enhancement, and edge detection, the calibration of the human eye position is completed, so that the two eyes are on the same horizontal line, which is convenient for subsequent portrait recognition;

[0014] Furthermore, all the images collected by the image acquisition unit of this application for a single individual at time t are used as the original data for subsequent recognition. Therefore, the sample picture set is the set of original images within time T. Compared with the prior art, operations such as flipping and cropping required for a single image acquisition by the image acquisition unit are omitted at the acquisition end, shortening the recognition time of the overall recognition system and improving the recognition efficiency;

[0015] A further improvement of the technical solution of the present invention lies in that all the extracted face samples are subjected to Sobel edge extraction, including:

[0016] Calculate the average value of the gray levels of the vector set x = [x0, x1,..., x n-1 of the sample image set to obtain the average gray level and the variance of the gray level distribution Furthermore, through Sobel edge extraction 0 ≤ i ≤ n, where x i is the gray level value of any pixel point in the sample image set. Finally, the gray level matrix T of the required original template for eyeball recognition and the face template is obtained. The area with the largest gray level in the original template for eyeball recognition is the position of the human eye.

[0017] Furthermore, the purpose of the preliminary recognition of this application for the special part of the human eye is that the human eye is the area with the largest gray level in the collected face gray image. Therefore, using the human eye as the preliminary positioning can quickly screen out the pictures in the sample picture set that do not capture the entire front face due to environmental factors such as angle and light. During the process of feedback of the recognition failure result to the image acquisition unit, when the number of recognition failure results exceeds the threshold, the image acquisition unit needs to re-collect or give an upper-end alarm.

[0018] A further improvement of the technical solution of the present invention is that, furthermore, after the eye recognition result determines the position of the human eye, in any picture, all image blocks of possible positions and possible sizes need to become candidate face R. Normalize R, measure the distance D between R and the vector T. If D is less than the pre-set value θ, then R is considered to be a face; otherwise, R is not a face. Then the vector T is replaced by the eye grayscale area that can best prove that the collected picture is a frontal face, that is, the vector T is the grayscale matrix T.

[0019] Screen the pictures in the input sample picture set that cannot be initially recognized according to the eye recognition result, including: the grayscale matrix T = {t ij}(i = 0, 1..., m - 1; j = 0, 1..., n - 1) is input into the image set matrix R = {r ij}(i = 0, 1..., m - 1; j = 0, 1..., n - 1) through the image acquisition unit, and calculate the distance d between any image in the image set and the human eye through the Euclidean distance calculation formula If d is within the pre-set threshold range, it means that the picture in the input image set matrix R is a face picture.

[0020] A further improvement of the technical solution of the present invention is that the average value of the image set feature vectors is calculated, including:

[0021] The obtained picture set feature vector x i = [x1, x2,..., x m , after The average value calculation gives the average value of the picture set feature vectors And the deviation set between the picture set feature vector x i and is in the form of matrix D. Compare the average value of the picture set feature vectors with the feature vectors of the built face database, and the deviation set matrix D is used as the credibility recognition threshold.

[0022] A further improvement of the technical solution of the present invention is that after the recognition unit provides the LBP features of each face image through the training model saving function, the recognition unit correspondingly records the feature data and labels of each face in the sample picture set and stores them in the face database.

[0023] A further improvement of the technical solution of the present invention is that the leading classifier in the recognition unit judges according to the similarity degree between the average value feature vector and the feature vectors in the built face database. The similarity degree judgment is calculated by the Chi-square statistical method. If the similarity degree exceeds the judgment credibility recognition threshold, it is considered an identification failure.

[0024] A further improvement of the technical solution of the present invention lies in that the image acquisition unit processes the face photos based on Dssd_inception_V4_coco to obtain the required sample picture set.

[0025] Beneficial effects

[0026] Compared with the prior art, the beneficial effects of the present invention are as follows: all the images collected by the image acquisition unit for a single individual at time t are used as the original data for subsequent recognition. The basic eyeball recognition template unit completes the preliminary recognition of the picture set, automatically filters out the pictures that cannot be preliminarily recognized in the input sample picture set, and introduces the obtained average feature vector of the filtered picture set into the nearest neighbor classifier to compare with the feature vectors of the established face database, realizing the portrait recognition of the picture set type. On the premise of realizing directional recognition, the rapid recognition of a large number of images belonging to a single face is completed, enabling the system to cope with random recognition tests and enabling the system to complete the recognition task in most scenarios;

[0027] All the images collected by the image acquisition unit for a single individual at time t are used as the original data for subsequent recognition. Therefore, the sample picture set is the set of original images within time T. Compared with the prior art, operations such as flipping and cropping necessary for single-image acquisition by the image acquisition unit are omitted at the acquisition end, shortening the recognition time of the overall recognition system and improving the recognition efficiency. Brief description of the drawings

[0028] Figure 1 It is a schematic diagram of the overall structure of a computer portrait recognition system;

[0029] Figure 2 It is a schematic diagram of the process of portrait recognition in a computer portrait recognition system. Detailed implementation manners

[0030] Hereinafter, various exemplary embodiments, features and aspects of the present application will be described in detail with reference to the drawings. The same reference numerals in the drawings denote elements having the same or similar functions. Although various aspects of the embodiments are shown in the drawings, the drawings do not have to be drawn to scale unless otherwise specified.

[0031] The special term "exemplary" here means "serving as an example, an embodiment or an illustration". Any embodiment described as "exemplary" here does not have to be construed as superior to or better than other embodiments.

[0032] In addition, for a better illustration of the present application, numerous specific details are given in the following specific embodiments. Those skilled in the art should understand that the present application can also be implemented without certain specific details. In some instances, methods, means, and components well-known to those skilled in the art are not described in detail so as to highlight the gist of the present application.

[0033] The present invention provides a computer human face recognition system, which includes:

[0034] An image acquisition unit; it acquires a face photo based on an optical sensing device, takes the face photo as the input of the computer, and completes the basic processing of the input image based on digital image processing to obtain a required sample picture set;

[0035] An eyeball basic recognition template unit; based on the acquired sample picture set, selects the face area as a face sample, performs Sobel edge extraction on all the extracted face samples, takes the average of the gray levels of the obtained edge image and transforms it to the required size as the original template for eyeball recognition, screens out the pictures in the input sample picture set that cannot be initially recognized according to the eyeball recognition result, and feeds the recognition failure result back to the image acquisition unit through the computer;

[0036] A recognition unit; the recognition unit completes the LBP feature extraction of the sample picture set based on a feature extraction subunit, obtains the image set feature vector of the sample picture set, calculates the average value of the image set feature vector, and introduces the obtained average value feature vector into a nearest neighbor classifier to compare it with the feature vectors of the established face database.

[0037] Embodiment 1, where the basic processing of the input image completed based on digital image processing includes: quantifying and sampling the spatial coordinates of the continuous input image function based on a computer to obtain a digital image set of the input image set;

[0038] Based on the computer, the pixels in the obtained digital image are composed in different ways, and noise filtering, image sharpening, contrast enhancement, and edge detection of the digital image are performed to obtain a required sample picture set;

[0039] Where the optical sensing device is, for example, a USB camera, etc., and a Linux system and V4L2 are used in the computer system to provide an image acquisition interface. After noise filtering, image sharpening, contrast enhancement, and edge detection, the image acquisition unit completes the calibration of the eye positions, making the two eyes on the same horizontal line, which is convenient for subsequent human face recognition;

[0040] Furthermore, all the images captured by the image acquisition unit of the present application for a single individual at time t are used as the original data for subsequent recognition. Therefore, the sample picture set is the set of original images within time T. Compared with the prior art, operations such as flipping and cropping required for single-image acquisition by the image acquisition unit are omitted at the acquisition end, shortening the recognition time of the overall recognition system and improving the recognition efficiency.

[0041] Embodiment 2, in which all the extracted face samples are subjected to Sobel edge extraction, including:

[0042] Calculate the average value of the gray levels of the vector set x = [x0, x1,..., x n-1 of the sample image set to obtain the average gray level and the variance of the gray level distribution , and then perform Sobel edge extraction 0 ≤ i ≤ n, where x i is the gray level value of any pixel point in the sample image set. Finally, obtain the gray matrix T of the required original template for eyeball recognition and the face template. The area with the maximum gray level in the original template for eyeball recognition is the position of the human eye.

[0043] Furthermore, the purpose of the present application for the preliminary recognition of the special part of the human eye is that the human eye is the area with the maximum gray level in the captured face gray image. Therefore, using the human eye as the preliminary positioning can quickly screen out the pictures in the sample picture set that do not capture the entire front face due to environmental factors such as angle and lighting. During the process of feedback of the recognition failure result to the image acquisition unit, when the number of recognition failure results exceeds the threshold, the image acquisition unit needs to re-acquire or give an upper-end alarm.

[0044] Furthermore, after determining the position of the human eye in the eyeball recognition result, in any picture, all possible image blocks of possible sizes at all possible positions need to become the candidate face R. Normalize R and measure the distance D between R and the vector T. If D is less than the pre-set value θ, then R is considered to be a face; otherwise, R is not a face. Then, the vector T is replaced by the eye gray area that can best prove that the captured picture is a front face, that is, the vector T is the gray matrix T;

[0045] Screen the pictures in the input sample picture set that cannot be initially recognized according to the eyeball recognition result, including: The gray matrix T = {t ij}(i = 0, 1..., m - 1; j = 0, 1..., n - 1) is input into the image set matrix R = {r ij}(i = 0, 1..., m - 1; j = 0, 1..., n - 1) by the image acquisition unit, and through the Euclidean distance calculation formula Calculate the distance d between any image in the image set and the human eye. If d is within the preset threshold range, it means that the picture in the input image set matrix R is a face picture.

[0046] Among them, calculating the average value of the image set feature vectors includes:

[0047] For the obtained picture set feature vector x i =[x1, x2,..., x m , after calculating the average value, the average value of the picture set feature vector is obtained And the deviation set between the picture set feature vector x i and is in the form of matrix D. Compare the average value of the picture set feature vector with the feature vectors of the built face database, and the deviation set matrix D is used as the credibility recognition threshold.

[0048] In Embodiment 3, after the recognition unit provides the LBP features of each face image, through the training model saving function, the recognition unit correspondingly records the feature data and labels of each face in the sample picture set and stores them in the face database.

[0049] Among them, the related functions of the training model are specifically:

[0050] / / Computes a LBPH model with images in src and

[0051] / / corresponding labels in labels, possibly preserving

[0052] / / old model data

[0053] void train(InputArrayOfArrays src, InputArray labels, bool perserveData);

[0054] After training through this model, a saving function is constructed to correspondingly record the feature data and labels of each face image. When there are no new portrait pictures added and there are no major environmental changes, the saved data can be directly used for recognition.

[0055] In the recognition unit, the leading classifier judges according to the similarity between the average value feature vector and the feature vectors in the built face database. The similarity judgment is calculated by the Chi-square statistical method. If the similarity exceeds the judgment credibility recognition threshold, it is considered an identification failure.

[0056] The image acquisition unit processes the face photos based on Dssd_inception_V4_coco to obtain the required sample picture set.

[0057] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device.

[0058] In specific implementation, the present application provides a computer storage medium and a corresponding data processing unit. Among them, the computer storage medium can store a computer program, and when the computer program is executed by the data processing unit, it can run the invention content of a computer portrait recognition system provided by the present invention and some or all of the steps in each embodiment. The storage medium can be a magnetic disk, an optical disk, a read-only memory or a random access memory, etc.

[0059] Those skilled in the art can clearly understand that the technical solutions in the embodiments of the present invention can be implemented by means of a computer program and its corresponding general hardware platform. Based on such an understanding, the essence of the technical solutions in the embodiments of the present invention, or the part that contributes to the prior art, can be embodied in the form of a computer program, that is, a software product. The computer program software product can be stored in the storage medium and includes several instructions for causing a device (which can be a personal computer, a server, a single-chip microcomputer, an MCU or a network device, etc.) including a data processing unit to execute the methods described in each embodiment or some parts of the embodiments of the present invention.

[0060] The present invention provides a computer portrait recognition system. There are many methods and ways to specifically implement this technical solution. The above is only the preferred implementation manner of the present invention. It should be pointed out that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention. Each component not clearly defined in this embodiment can be implemented by the prior art.

Claims

1. A computer portrait recognition system, characterized in that: The system comprises: Image acquisition unit: collects face photos based on optical sensing equipment, and the face photos are used as computer input. Based on digital image processing, basic processing of the input images is completed to obtain the required sample picture set; The basic eye recognition template unit: based on the acquired sample picture set, the face area is selected as the face sample, and all the extracted face samples are subjected to Sobel edge extraction, the grayscale of the obtained edge image is averaged and transformed to the required size as the original template for eye recognition, and the pictures that cannot be preliminarily recognized in the input sample picture set are screened according to the eye recognition results, and the recognition failure results are fed back to the image acquisition unit via the computer; Recognition unit: The recognition unit completes LBP feature extraction of the sample picture set based on the feature extraction subunit, obtains the image set feature vector of the sample picture set, calculates the average value of the image set feature vector, and introduces the obtained average feature vector into the neighbor classifier for comparison with the feature vector of the established face library.

2. A computer portrait recognition system according to claim 1, characterized in that: The basic processing of the input image is completed based on digital image processing, including: completing the quantization and sampling of the spatial coordinates of the continuous input image function based on the computer to obtain the digital image set of the input image set; The pixels in the obtained digital image are assembled in different ways based on a computer, and the digital image is subjected to noise filtering, image sharpening, contrast enhancement and edge detection to obtain the required sample image set.

3. A computer portrait recognition system according to claim 1, characterized in that: All extracted face samples are subjected to Sobel edge extraction, including: The vector set x of the sample image set is x = [x0, x1, ..., x n-1 ] Calculate the average value of grayscale and get the average grayscale value And the grayscale distribution variance Then use Sobel edge extraction where x i is the grayscale value of any pixel in the sample image set, and finally the grayscale matrix T of the required eye recognition original template and face template is obtained. The area with the largest grayscale in the eye recognition original template is the position of the human eye.

4. A computer portrait recognition system according to claim 3, characterized in that: According to the eyeball recognition results, the images that cannot be initially recognized in the input sample image set are filtered, including: gray matrix T = {t ij }(i=0,1...,m-1;j=0,1...,n-1) is input into the image set matrix R={r ij }(i=0,1...,m-1;j=0,1...,n-1), and calculate the distance by the Euclidean distance formula The distance d between any image in the image set and the human eye is calculated. If d is within a preset threshold range, it means that the image in the input image set matrix R is a face image.

5. A computer portrait recognition system according to claim 1, characterized in that: The average value of the feature vector of the image set is calculated, including: The obtained picture set feature vector x i =[x1,x2,...,x m ],go through The average value is calculated to get the average value of the feature vector of the picture set And the picture set feature vector x i and The deviation set between them is in the form of matrix D, and the average value of the feature vector of the picture set is Compared with the feature vector of the built face database, the deviation set matrix D is used as the credibility recognition threshold.

6. A computer portrait recognition system according to claim 1, characterized in that: The recognition unit provides the LBP features of each face image and saves the function through the training model. The recognition unit records the feature data and labels of each face in the sample picture set and stores them in the face library.

7. A computer portrait recognition system according to claim 5, characterized in that: The classifier in the recognition unit judges the similarity between the average feature vector and the feature vector in the built face database. The similarity judgment is calculated by the Chi square statistical method. If the similarity exceeds the judgment credibility recognition threshold, the recognition fails.

8. A computer portrait recognition system according to claim 1, characterized in that: The image acquisition unit performs data processing on the face photo based on Dssd_inception_V4_coco to obtain the required sample picture set.

Images