Automatic image processing method and system based on face features

Through the face detection algorithm and automation processing module based on deep learning, the problem of insufficient accuracy and insufficient automation capabilities of the face image processing system in the prior art in complex environments is solved, and efficient and personalized face images are achieved Processing improves processing accuracy and speed.

CN118982855AInactive Publication Date: 2024-11-19SHANGHAI WANBI QIANMO TECHNOLOGY CO LTD
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
CN202411067528.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-06
Publication Date
2024-11-19
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The facial image processing system in the prior art is insufficient in the accuracy of complex backgrounds or posture changes, the facial key point detection and calibration technology is not accurate, and the processing ability is lacking in automation and personalization, resulting in low processing efficiency and poor consistency.

Method used

Using a face detection algorithm based on deep learning, the candidate face area is generated through the convolutional neural network and the regional suggestion network, and face detection and key point calibration are carried out in combination with the non-maximum suppression algorithm, and face map reconstruction and clarification are used. Processing, fusion and beauty modules enable automated image processing.

Benefits of technology

Significantly improves the accuracy and robustness of face detection, the accuracy of facial key point detection, reduces the need for manual adjustment, improves processing efficiency and personalization, optimizes the processing speed and real-time of the system, and meets large-scale Efficiency requirements for image processing.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118982855A_ABST
    Figure CN118982855A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of face feature recognition, in particular to an automatic image processing method and system based on face features, and the method comprises the steps: inputting a to-be-processed image; a face detection module is used for automatically detecting and positioning a face area of the image; automatically calibrating 68 key points of a human face by using a face key point detection module; the face map reconstruction module is utilized to synthesize a new face map according to mesh information of a reference face and a target face, through a face detection algorithm, the accuracy and robustness of face detection in a complex environment are remarkably improved, the automatic detection and calibration technology of face key points is more accurate, and the face detection efficiency is improved. High accuracy can be kept even in complex scenes such as shielding and expression change, the system not only reduces the requirement of manual adjustment, but also improves the processing efficiency through automatic face image fusion and beautification processing, the processing speed and real-time performance of the whole system are optimized, and the user experience is improved. And the efficiency requirement of large-scale image processing can be met.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of facial feature recognition, and in particular to an automatic image processing method and system based on facial features. Background Art

[0002] With the rapid growth of digital image and video content, face image processing technology has become increasingly important in many fields such as security monitoring, social media, entertainment, and identity authentication. Traditional face image processing methods mainly rely on manual editing or simple image processing algorithms, which have problems such as low efficiency, poor consistency, and inability to process large amounts of data.

[0003] The deep learning-based method can automatically learn complex feature representations from a large amount of data to achieve high-precision face detection, recognition and editing. However, the face image processing system in the prior art still has the following deficiencies: First, the face detection algorithm in the prior art is not accurate enough when dealing with complex backgrounds or posture changes, resulting in a decrease in the accuracy of subsequent processing steps; second, the automatic detection and calibration technology of facial key points is not yet mature, especially when dealing with complex situations such as occlusion and expression changes, it is difficult to achieve ideal accuracy; third, the face image fusion and beautification processing in the prior art often requires a lot of manual adjustments, lacking automation and personalized processing capabilities. Summary of the invention

[0004] In view of the above problems or problems existing in the prior art, the present invention is proposed.

[0005] Therefore, the object of the present invention is to provide an automatic image processing method based on facial features.

[0006] In order to solve the above technical problems, the present invention provides the following technical solution: comprising the following steps:

[0007] Input the image to be processed;

[0008] Use the face detection module to automatically detect and locate the face area of ​​the image;

[0009] The facial key point detection module is used to automatically calibrate the 68 key points of the face;

[0010] The face map reconstruction module is used to synthesize a new face map based on the mesh information of the reference face and the target face;

[0011] The synthesized face map is processed into high definition using the face clearing processing module;

[0012] Use the face fusion module to naturally fuse the high-definition face maps;

[0013] Use the face beautification module and the face skin beautification module to beautify and skin beautify the fused face;

[0014] Output the image processed by the above steps.

[0015] As a preferred solution of the automatic image processing method based on facial features of the present invention, the face detection module adopts a face detection algorithm based on deep learning and is implemented by the following steps: input an image containing an unknown number and position of faces; use a convolutional neural network to extract features of the input image, wherein the convolutional neural network includes multiple convolutional layers and pooling layers for extracting image features layer by layer; on the basis of the convolutional neural network, a region proposal network (RPN) is used to generate candidate face regions, and the RPN searches for potential face regions at different scales through a sliding window; a fully connected layer is applied to the candidate regions generated by the RPN to perform face classification and bounding box regression; a non-maximum suppression algorithm is applied to remove overlapping face detection frames to retain the best detection results; and the final face position and size information is output.

[0016] As a preferred solution of the automatic image processing method based on facial features of the present invention, the convolutional neural network includes but is not limited to the following layers: multiple convolutional layers, used to apply convolution kernels for feature extraction; activation layers, such as ReLU layers, used to introduce nonlinearity and enhance the expression ability of the network; pooling layers, used to reduce the spatial dimension of the features while increasing the invariance to image displacement.

[0017] As a preferred solution of the automatic image processing method based on facial features of the present invention, the bounding box regression adopts the following formula:

[0018] [x,y,w,h]=[x p ,y p ,w p ,h p ]+[dx,dy][dw,dh]

[0019] Among them, [x, y, w, h] refers to the predicted box after transformation, [x p ,y p ,w p ,h p ] refers to the initial proposed face candidate region, [dx,dy] refers to the offset of the center point of the bounding box, and [dw,dh] refers to the scaling factor of the width and height of the bounding box.

[0020] As a preferred solution of the automatic image processing method based on facial features of the present invention, the bounding box regression formula also includes:

[0021] x=x p +dx;

[0022] y=y p +dy;

[0023] w=w p *e dw ;

[0024] h=h p *e dh ;;

[0025] Here, e is the base of natural logarithms and is used to express the exponential form of the scaling factor.

[0026] As a preferred solution of the automatic image processing method based on facial features of the present invention, the non-maximum suppression algorithm comprises the following steps:

[0027] Step 1: For each category, sort the detection boxes according to their confidence scores.

[0028] Step 2: Select the detection box with the highest confidence as the reference box.

[0029] Step 3: Calculate the intersection over union (IoU) between other detection boxes and the reference box.

[0030] Step 4: If the IoU between the detection box and the reference box is higher than a preset threshold, the detection box is suppressed (i.e., removed from the candidate list);

[0031] In step 5, continue to select the next unsuppressed detection box with the highest confidence as the new reference box, and repeat steps 3 and 4 until all detection boxes have been considered.

[0032] As a preferred solution of the automatic image processing method based on facial features of the present invention, the calculation formula of IoU is:

[0033]

[0034] Among them, AOI refers to Area of ​​Intersection, which is the area of ​​the overlapping area of ​​two bounding boxes, and AOU refers to Area of ​​Union, which is the area of ​​the union of two bounding boxes, which is equal to the sum of their respective areas minus the area of ​​the overlapping area.

[0035] As a preferred solution of the automatic image processing method based on facial features of the present invention, the IoU is calculated as follows:

[0036] If there are two bounding boxes A and B, their coordinates are (x A ,y A , wA ,h A ) and (x B ,y B , w B ,h B ), where w and h represent the width and height of the bounding box, respectively.

[0037] First calculate the coordinates of the intersection area of ​​the two bounding boxes:

[0038] Upper left corner coordinate, x min =max(x A , x B );

[0039] Coordinate of the lower right corner, y min =min(y A +h A ,y B +h B );

[0040] Width intersect =max(0,y min -x min );

[0041] Height intersect =max(0, x min -y min );

[0042] Then calculate the intersection area, AOI = w intersec *h intersect ;

[0043] Then calculate the area of ​​each of the two bounding boxes,

[0044] S A =w A *h A ;

[0045] S B =w B *h B ;

[0046] Then calculate the area of ​​the union region, AOU = S A +S B -AOI;

[0047] Finally, calculate IoU.

[0048] As a preferred solution of the automatic image processing method based on facial features of the present invention, in practical application, if w intersect or intersectIf it is a negative number, that is, the two bounding boxes do not overlap, the IoU will be zero, indicating that the two bounding boxes have no or only a small overlap.

[0049] As a preferred solution of the automatic image processing method based on facial features of the present invention, wherein: the processor implements the steps of the above-mentioned automatic image processing method based on facial features when executing the computer program.

[0050] The beneficial effects of the automated image processing method based on facial features of the present invention are as follows: through the face detection algorithm, the accuracy and robustness of face detection in complex environments are significantly improved; secondly, the automatic detection and calibration technology of facial key points is more precise, and can maintain high accuracy even in complex scenes such as occlusion and expression changes; in addition, the system uses automated face image fusion and beautification processing, which not only reduces the need for manual adjustment, but also improves the processing efficiency and personalization; the processing speed and real-time performance of the overall system are optimized, which can meet the efficiency requirements of large-scale image processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0052] Figure 1 A diagram of a computer device for an automated image processing method based on facial features. DETAILED DESCRIPTION

[0053] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0054] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0055] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0056] Example 1, reference Figure 1, which is the first embodiment of the present invention, and provides an automatic image processing method based on facial features, which includes the following steps: inputting an image to be processed; using a face detection module to automatically detect and locate the face area of ​​the image; using a facial key point detection module to automatically calibrate 68 key points of the face; using a face map reconstruction module to synthesize a new face map according to the mesh information of a reference face and a target face; using a face sharpening processing module to perform high-definition processing on the synthesized face map; using a face fusion module to naturally fuse the high-definition processed face map; using a face beauty module and a face skin beauty module to perform beauty and skin beauty processing on the fused face; and outputting the image processed by the above steps.

[0057] Specifically, the face detection module adopts a face detection algorithm based on deep learning, which is implemented by the following steps: input an image containing an unknown number and position of faces; use a convolutional neural network to extract features of the input image, wherein the convolutional neural network includes multiple convolutional layers and pooling layers for extracting image features layer by layer; based on the convolutional neural network, use a region proposal network (RPN) to generate candidate face regions, and the RPN searches for potential face regions at different scales through a sliding window; apply a fully connected layer to the candidate regions generated by the RPN to perform face classification and bounding box regression; apply a non-maximum suppression algorithm to remove overlapping face detection frames and retain the best detection results; and output the final face position and size information.

[0058] Furthermore, the convolutional neural network includes but is not limited to the following layers: multiple convolutional layers, used to apply convolution kernels for feature extraction; activation layers, such as ReLU layers, used to introduce nonlinearity and enhance the expressive power of the network; pooling layers, used to reduce the spatial dimension of features while increasing invariance to image displacement.

[0059] Among them, the bounding box regression adopts the following formula, [x, y, w, h] = [xp, yp, wp, hp] + [dx, dy] [dw, dh], wherein [x, y, w, h] refers to the prediction box obtained after the transformation, [xp, yp, wp, hp] refers to the initially proposed face candidate area, [dx, dy] refers to the offset of the center point of the bounding box, and [dw, dh] refers to the scaling factor of the width and height of the bounding box.

[0060] 1. Preferably, the bounding box regression formula also includes:

[0061] x=x p +dx;

[0062] y=y p +dy;

[0063] w=w p *e dw ;

[0064] h=h p *e dh ;;

[0065] Here, e is the base of natural logarithms and is used to express the exponential form of the scaling factor.

[0066] It should be noted that the non-maximum suppression algorithm has the following steps: Step 1, for each category, sort the detection boxes according to their confidence scores. Step 2, select the detection box with the highest confidence as the reference box. Step 3, calculate the intersection over union (IoU) of other detection boxes and the reference box. Step 4, if the IoU between the detection box and the reference box is higher than a preset threshold, suppress the detection box (i.e., remove it from the candidate list); Step 5, continue to select the next detection box with the highest confidence that has not been suppressed as the new reference box, and repeat steps 3 and 4 until all detection boxes have been considered.

[0067] When used, the calculation formula of the IoU is,

[0068]

[0069] Among them, AOI refers to Area of ​​Intersection, which is the area of ​​the overlapping area of ​​two bounding boxes, and AOU refers to Area of ​​Union, which is the area of ​​the union of two bounding boxes, which is equal to the sum of their respective areas minus the area of ​​the overlapping area.

[0070] In particular, the IoU is calculated as follows,

[0071] If there are two bounding boxes A and B, their coordinates are (x A ,y A , w A ,h A ) and (x B ,y B , w B ,h B ), where w and h represent the width and height of the bounding box, respectively.

[0072] First calculate the coordinates of the intersection area of ​​the two bounding boxes:

[0073] Upper left corner coordinate, x min =max(x A , x B );

[0074] Coordinate of the lower right corner, y min =min(y A+h A ,y B +h B );

[0075] Width intersect =max(0,y min -x min );

[0076] Height intersect =max(0, x min -y min );

[0077] Then calculate the intersection area, AOI = w intersec *h intersect ;

[0078] Then calculate the area of ​​each of the two bounding boxes,

[0079] S A =w A *h A ;

[0080] S B =w B *h B ;

[0081] Then calculate the area of ​​the union region, AOU = S A +S B -AOI;

[0082] Finally, calculate IoU.

[0083] Furthermore, in practical applications, if w intersect or intersect If it is a negative number, that is, the two bounding boxes do not overlap, the IoU will be zero, indicating that the two bounding boxes have no or only a small overlap.

[0084] By combining the face detection module and the facial key point detection module, the present invention can more accurately locate the face position and calibrate the key points, thereby improving the precision and accuracy of face processing, avoiding the position deviation and inaccurate key point calibration problems that may exist in traditional methods, and using the face mapping reconstruction module and the face sharpening processing module, the present invention can reconstruct a new face image with reference facial information features, and improve the image quality through sharpening processing, making the processing result clearer and more realistic, avoiding the image blur and distortion in traditional processing methods, and through the application of the face fusion module, the present invention can perform re-fusion on the basis of the processed face image , reconstructing a more natural and realistic face picture, enhancing the similarity between the face in the photo and the reference face, avoiding the problem of unnatural transition in the traditional fusion method, and combining the key point information of the face, the face beautification module and the face skin beautification module of the present invention can achieve beauty effects such as face pulling, face thinning, and eye enlargement, and at the same time process the skin to achieve effects such as acne removal, scar removal, and whitening, thereby improving the beauty and attractiveness of the image, making the processing result more in line with the personalized needs of the user, having higher accuracy, quality and naturalness in the field of face processing, and providing users with a better face processing experience. Compared with the existing technology, it has obvious technical improvements and effect improvements.

[0085] In summary, firstly, by inputting the image to be processed, the system automatically starts the face detection module to quickly locate the face area in the image; then, the facial key point detection module accurately calibrates the facial feature points to provide necessary information for the three-dimensional modeling of the face; then, the face texture reconstruction module synthesizes a new texture based on the extracted feature information to achieve the initial reconstruction of the face; then, the face sharpening processing module optimizes the details of the reconstructed face to enhance the clarity and realism of the image; further, the face fusion module adopts advanced fusion technology to make the processed face more harmonious and natural with the original image; finally, the face beauty and skin beautification modules provide personalized beautification effects according to user needs, complete the entire automated face image processing process, and output high-quality processing results. The entire process does not require complex manual intervention, and realizes fully automated processing from detection to beautification, greatly improving processing speed and user satisfaction.

[0086] This embodiment also provides a computer device, which is suitable for the case of an automated image processing method based on facial features, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement all or part of the steps of the method described in the embodiment of the present invention as proposed in the above embodiment.

[0087] This embodiment also provides a storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the method in any optional implementation of the above embodiment is executed. Wherein, the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, referred to as EPROM), programmable read-only memory (Programmable Red-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0088] Importantly, it should be noted that the construction and arrangement of the present application shown in a plurality of different exemplary embodiments are only exemplary. Although only a few embodiments are described in detail in this disclosure, it should be readily understood by those who refer to this disclosure that many modifications are possible (e.g., the size, scale, structure, shape and proportion of various elements, and parameter values ​​(e.g., temperature, pressure, etc.), installation arrangement, use of materials, color, directional changes, etc.) without substantially departing from the novel teachings and advantages of the subject matter described in the application. For example, the element shown as integrally formed can be composed of multiple parts or elements, the position of the element can be inverted or otherwise changed, and the nature or number or position of the discrete element can be changed or changed. Therefore, all such modifications are intended to be included in the scope of the present invention. The order or sequence of any process or method steps can be changed or reordered according to alternative embodiments. In the claims, any "device plus function" clause is intended to cover the structure of performing the function described herein, and is not only structurally equivalent but also equivalent structure. Without departing from the scope of the present invention, other replacements, modifications, changes and omissions can be made in the design, operating conditions and arrangement of the exemplary embodiments. Therefore, the invention is not limited to a specific embodiment, but extends to several modifications still falling within the scope of the appended claims.

[0089] Additionally, in order to provide a concise description of exemplary embodiments, all features of an actual embodiment (ie, those features that are not relevant to the best mode presently contemplated for carrying out the invention or those that are not relevant to implementing the invention) may not be described.

[0090] It will be appreciated that in the development of any actual implementation, as in any engineering or design project, numerous implementation-specific decisions may be made. Such a development effort may be complex and time-consuming, but will be a routine task of design, fabrication, and production for those of ordinary skill having the benefit of this disclosure without undue experimentation.

[0091] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An automated image processing method based on facial features, characterized in that: The following steps are included: Input the image to be processed; Use the face detection module to automatically detect and locate the face area of ​​the image; The facial key point detection module is used to automatically calibrate the 68 key points of the face; The face map reconstruction module is used to synthesize a new face map based on the mesh information of the reference face and the target face; The synthesized face map is processed into high definition using the face clearing processing module; Use the face fusion module to naturally fuse the high-definition face maps; Use the face beautification module and the face skin beautification module to beautify and skin beautify the fused face; Output the image processed by the above steps.

2. The automatic image processing method based on facial features as claimed in claim 1, characterized in that: The face detection module adopts a face detection algorithm based on deep learning, which is implemented by the following steps: Input an image containing unknown number and position of faces; Using a convolutional neural network to extract features from an input image, wherein the convolutional neural network includes a plurality of convolutional layers and pooling layers for extracting image features layer by layer; Based on the convolutional neural network, the Region Proposal Network (RPN) is used to generate candidate face regions. The RPN searches for potential face regions at different scales through a sliding window. Apply a fully connected layer to the candidate regions generated by RPN for face classification and bounding box regression; Apply non-maximum suppression algorithm to remove overlapping face detection frames and retain the best detection results; Output the final face position and size information.

3. The automatic image processing method based on facial features as claimed in claim 2, characterized in that: The convolutional neural network includes but is not limited to the following layers: Multiple convolutional layers for applying convolution kernels for feature extraction; Activation layers, such as ReLU layers, are used to introduce nonlinearity and enhance the expressiveness of the network; Pooling layer is used to reduce the spatial dimension of features while increasing invariance to image displacement.

4. The automatic image processing method based on facial features as claimed in claim 3, characterized in that: The bounding box regression adopts the following formula, [x,y,w,h]=[x p ,y p ,w p ,h p ]+[dx,dy][dw,dh] Among them, [x, y, w, h] refers to the predicted box after transformation, [x p ,y p ,w p ,h p ] refers to the initial proposed face candidate region, [dx,dy] refers to the offset of the center point of the bounding box, and [dw,dh] refers to the scaling factor of the width and height of the bounding box.

5. The automatic image processing method based on facial features as claimed in claim 4, characterized in that: The bounding box regression formula also includes, x=x p +dx; y=y p +dy: w=w p *have been dw ; h=h p *e dh ;; Here, e is the base of natural logarithms and is used to express the exponential form of the scaling factor.

6. The automatic image processing method based on facial features as claimed in claim 4 or 5, characterized in that: The non-maximum suppression algorithm has the following steps: Step 1: For each category, sort the detection boxes according to their confidence scores; Step 2: Select the detection frame with the highest confidence as the reference frame; Step 3: Calculate the intersection over union (IoU) between other detection boxes and the reference box. Step 4: If the IoU between the detection box and the reference box is higher than a preset threshold, the detection box is suppressed (i.e., removed from the candidate list); In step 5, continue to select the next unsuppressed detection box with the highest confidence as the new reference box, and repeat steps 3 and 4 until all detection boxes have been considered.

7. The automatic image processing method based on facial features as claimed in claim 6, characterized in that: The calculation formula of IoU is: Among them, AOI refers to Area of ​​Intersection, which is the area of ​​the overlapping area of ​​two bounding boxes, and AOU refers to Area of ​​Union, which is the area of ​​the union of two bounding boxes, which is equal to the sum of their respective areas minus the area of ​​the overlapping area.

8. The automatic image processing method based on facial features as claimed in claim 1, characterized in that: The IoU is calculated as follows, If there are two bounding boxes A and B, their coordinates are (x A ,y A , w A ,h A ) and (x B ,y B , w B ,h B ), where w and h represent the width and height of the bounding box, respectively. First calculate the coordinates of the intersection area of ​​the two bounding boxes: Upper left corner coordinate, x min =max(x A , x B ); Coordinate of the lower right corner, y min =min(y A +h A ,y B +h B ); Width intersect =max(0,y min -x min ); Height intersect =max(0, x min -y min ); Then calculate the intersection area, AOI = w intersec *h intersect ; Then calculate the area of ​​each of the two bounding boxes, S A =w A *h A ; S B =w B *h B ; Then calculate the area of ​​the union region, AOU = S A +S B -AOI; Finally, calculate IoU.

9. The automatic image processing method based on facial features as claimed in claim 8, characterized in that: In practical applications, if w intersect or intersect If it is a negative number, that is, the two bounding boxes do not overlap, the IoU will be zero, indicating that the two bounding boxes have no or only a small overlap.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of any one of the automatic image processing methods based on facial features in claims 1 to 9 are implemented.

Citation Information

Patent Citations

  • Single photograph-based fast face modeling method

    CN107316340A

  • Classroom attendance checking method and equipment based on neural network, storage medium and classroom attendance checking device based on neural network

    CN110097003A

  • Three-dimensional face model map fusion method and computer processing equipment

    CN110232730A

  • Multi-student individual segmentation and state autonomous identification method based on deep learning

    CN111507227A

  • Face detection and counting method and system based on video optical flow information

    CN114663476A