Face feature recognition security control method and system based on machine learning

Abstract

The present invention relates to the field of security technology, and discloses a facial feature recognition security control method based on machine learning, comprising the following steps: obtaining a sampled video including a target face; setting a dynamic mark at a designated position of the target face; obtaining a standard picture corresponding to the sampled video according to the dynamic mark; judging the security level of the target face according to the standard picture, and adjusting and updating the designated position according to the security level. The dynamic mark that can be changed determines that the standard picture will also change accordingly, which increases the probability of the second security level judgment changing. External users cannot accurately know the change logic of the dynamic mark, nor can they accurately make corresponding disguises, thereby improving the overall security performance.

Description

Technical Field

[0001] The present invention relates to the field of security technology, and in particular to a facial feature recognition security control method and system based on machine learning. Background Art

[0002] At present, the access control security related systems mainly include systems using IC card, image recognition, fingerprint recognition and other identification methods, but their security is relatively low. Images, fingerprints and IC cards can be stolen by humans, thus easily breaking through the security system. With the further maturity of technology and the improvement of social recognition, face recognition technology will be applied in more fields.

[0003] Face recognition technology is based on human facial features to judge the input face image or video, and many important breakthroughs have been made. Face recognition technology is based on human facial features to judge the input face image or video. The system input is generally one or a series of face images with unidentified identities, and the system output is a series of similarity scores, which can be used to determine whether to allow access.

[0004] For example, training samples are used to train the neural network model, and the trained neural network model is used as a recognition module. The collected face images are then sent to the recognition module for processing. This can quickly determine whether the identity meets the requirements and output a probability value that meets a certain level of security. The higher the probability value, the higher the confidence level of the security level. When the confidence level exceeds a certain value, the security level judgment can be output.

[0005] However, although the existing face recognition can achieve a high degree of recognition accuracy, there are still problems in the system input stage that may cause the recognition accuracy to decrease. For example, outsiders who know in advance that the logic of recognition judgment is the overall facial contour will deliberately use relevant means to modify the facial contour to evade security review, so as to deceive the automatic recognition of the neural network model. After all, the training samples of the neural network model are limited, and the entire face is used as a sample for training, which may lead to misjudgment of the security level. Summary of the invention

[0006] The purpose of the present invention is to provide a facial feature recognition security control method and system based on machine learning to solve the following technical problems:

[0007] How to provide a more accurate and secure facial feature recognition security control method.

[0008] The purpose of the present invention can be achieved through the following technical solutions:

[0009] The facial feature recognition security control method based on machine learning includes the following steps:

[0010] S100, obtaining a sample video including a target face;

[0011] S200, setting a dynamic mark at a designated position of the target face;

[0012] S300, acquiring a standard picture corresponding to the sampled video according to the dynamic identifier;

[0013] S400, determining the security level of the target face according to the standard image, and adjusting and updating the designated position according to the security level.

[0014] Through the above technical solution, the dynamic mark set on the target face can be changed according to the security level obtained by subsequent judgment, so as to further re-acquire and identify the standard image, and take the security level judged subsequently as the real judgment result. In other words, the dynamic mark that can be changed determines that the standard image will also change accordingly, thereby increasing the probability of changing the second security level judgment. External users cannot accurately know the changing logic of the dynamic mark, and therefore cannot accurately make corresponding disguises, thereby improving the overall security performance.

[0015] As a further solution of the present invention: step S200 includes:

[0016] Acquire a sample frame in the sample video; wherein the sample video includes a facial guidance frame, and the facial guidance frame is used to guide the user to adjust the facial position;

[0017] When the user's facial position is consistent with the facial guidance frame, activating the dynamic identification;

[0018] The timestamp of the sampling frame is included in the start time of the dynamic identifier.

[0019] Through the above technical solution, the sampling video can only be shot after the user adjusts the face into the face guidance frame, which can greatly reduce the burden of recognition and reduce the complexity. The acquisition time of the sampling frame and the start time of the dynamic mark are consistent, which can reduce the impact of the dynamic mark on the user himself and be more humane.

[0020] As a further solution of the present invention: step S200 includes:

[0021] The designated positions include positions between adjacent part types, and the part types include eyebrows, nose bridge, eyes, and lips.

[0022] As a further solution of the present invention: step S300 includes:

[0023] For any of the sampling frames, selecting a preset range encircled by the dynamic mark in the sampling frame as a selected picture;

[0024] Combining the selected images corresponding to all the sampling frames within a preset time into a recognition image according to a synthesis rule;

[0025] The identified image is processed into a standard image according to preset rules.

[0026] Through the above technical solution, the synthesized recognition photo is involved in all sampling frames within a preset time, so as to reduce the pixel noise in a single sampling frame, and then the recognition picture is processed into a standard picture to facilitate subsequent system input.

[0027] As a further solution of the present invention: the synthesis rule includes:

[0028] Merging the selected images according to the maximum overlap of identical pixels;

[0029] The same pixels remain unchanged, and the middle value of the different pixels is taken to obtain the identified image.

[0030] Through the above technical solution, merging according to the maximum overlap of the same pixels can ensure that the synthesized recognition image will not be blurred or distorted. Taking the middle value of different pixels to merge can weaken the influence of external light to a certain extent and improve the accuracy of subsequent recognition judgment.

[0031] As a further solution of the present invention: the preset rules include:

[0032] Take the standard framework;

[0033] placing the identification picture in the standard frame at a preset angle;

[0034] The positions in the standard frame that do not overlap with the identified picture are filled with a solid color to obtain a standard picture.

[0035] Through the above technical solution, the preset picture is placed in the standard frame, and the standard picture formed after filling can be recognized in a unified form, which helps to reduce the complexity of recognition. In addition, the adjustment of the preset angle can make the relevant modules of the recognition process remain unified during training, thereby further reducing the recognition pressure.

[0036] As a further solution of the present invention: a facial feature recognition security control system based on machine learning, comprising:

[0037] A sampling module, used to obtain a sample video including a target face;

[0038] A feature marking module, connected to the sampling module, for setting a dynamic mark at a designated position of the target face;

[0039] A processing module, which obtains a standard picture corresponding to the sampled video according to the dynamic identifier;

[0040] The recognition output module is used to determine the security level of the target face according to the standard image, and adjust and update the designated position according to the security level.

[0041] Wherein, the recognition output module is a trained artificial neural network.

[0042] The part types include eyebrows, nose bridge, eyes, and lips.

[0043] Beneficial effects of the present invention:

[0044] (1) The dynamic mark set on the target face of the present invention can be changed according to the security level obtained by subsequent judgment, so as to further re-acquire and identify the standard image, and take the security level judged subsequently as the real judgment result. In other words, the dynamic mark that can be changed determines that the standard image will also change accordingly, thereby increasing the probability of the second security level judgment changing. External users cannot accurately know the changing logic of the dynamic mark, and therefore cannot accurately make corresponding disguises, thereby improving the overall security performance;

[0045] (2) In the present invention, the sampling video can be shot only after the user adjusts the face into the face guidance frame, which can greatly reduce the burden of recognition and reduce the complexity. The acquisition time of the sampling frame and the start time of the dynamic mark can be consistent, which can reduce the impact of the dynamic mark on the user himself, making it more humane.

[0046] (3) In the present invention, merging is performed according to the maximum overlap of identical pixels, which can ensure that the synthesized recognition image will not be blurred or distorted. Merging the median values ​​of different pixels can weaken the influence of external light to a certain extent, thereby improving the accuracy of subsequent recognition judgment. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The present invention will be further described below in conjunction with the accompanying drawings.

[0048] Figure 1 This is a facial feature recognition security control method based on machine learning in the present invention. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] See also Figure 1 As shown, the present invention is a facial feature recognition security control method based on machine learning, comprising the following steps:

[0051] S100, obtaining a sample video including a target face;

[0052] S200, setting a dynamic mark at a designated position of the target face;

[0053] S300, acquiring a standard image corresponding to the sampled video according to the dynamic identifier;

[0054] S400, judging the safety level of the target face according to the standard image, and adjusting and updating the designated position according to the safety level.

[0055] Through the above technical solution, the dynamic mark set on the target face can be changed according to the security level obtained by subsequent judgment, so as to further re-acquire and identify the standard image, and take the security level judged subsequently as the real judgment result. In other words, the dynamic mark that can be changed determines that the standard image will also change accordingly, thereby increasing the probability of changing the second security level judgment. External users cannot accurately know the changing logic of the dynamic mark, and therefore cannot accurately make corresponding disguises, thereby improving the overall security performance.

[0056] As a further solution of the present invention: step S200 includes:

[0057] Acquire a sample frame in a sample video; wherein the sample video includes a facial guidance frame, and the facial guidance frame is used to guide the user to adjust the facial position;

[0058] When the user's face position is consistent with the face guide frame, the dynamic identification is activated;

[0059] The timestamp of the sampling frame is included in the start time of the dynamic mark.

[0060] Through the above technical solution, the sampling video can only be shot after the user adjusts the face into the face guidance frame, which can greatly reduce the burden of recognition and reduce the complexity. The acquisition time of the sampling frame and the start time of the dynamic mark are consistent, which can reduce the impact of the dynamic mark on the user himself and be more humane.

[0061] When in use, the user can see a screen containing his or her own face, in which the above-mentioned face guide frame is set. When the user adjusts the face into the face guide frame, the sample video starts recording, and the dynamic mark starts setting at the same time. In this embodiment, the dynamic mark can be irradiated to the user's face in the form of infrared irradiation. Of course, in order to avoid damage to the user's eyesight, the irradiation time needs to be greatly shortened. Therefore, the sampling frame is acquired within the time of starting the dynamic mark.

[0062] As a further solution of the present invention: step S200 includes:

[0063] The designated positions include positions between adjacent part types, and the part types include eyebrows, a nose bridge, eyes, and lips.

[0064] As a further solution of the present invention: step S300 includes:

[0065] For any sampling frame, a preset range encircled by a dynamic mark in the sampling frame is selected as a selected picture;

[0066] The selected images corresponding to all sampling frames within a preset time are synthesized into a recognition image according to a synthesis rule;

[0067] Process the recognized images into standard images according to preset rules.

[0068] Through the above technical solution, the synthesized recognition photo is involved in all sampling frames within a preset time, so as to reduce the pixel noise in a single sampling frame, and then the recognition picture is processed into a standard picture to facilitate subsequent system input.

[0069] In this embodiment, taking the recognition of the eye area as an example, the dynamic mark can be projected to the user's eye position, and then 4 sampling frames are obtained within 0.01 seconds, thereby obtaining 4 selected pictures corresponding to the sampling frames respectively, and then the 4 selected pictures are merged to obtain a recognition picture.

[0070] As a further solution of the present invention: the synthesis rule includes:

[0071] Merge the selected images according to the maximum overlap of the same pixels;

[0072] The same pixels remain unchanged, and the middle value of the different pixels is taken to obtain the recognized image.

[0073] Through the above technical solution, merging according to the maximum overlap of the same pixels can ensure that the synthesized recognition image will not be blurred or distorted. Taking the middle value of different pixels to merge can weaken the influence of external light to a certain extent and improve the accuracy of subsequent recognition judgment.

[0074] As a further solution of the present invention: the preset rules include:

[0075] Take the standard framework;

[0076] Place the identification image in a standard frame at a preset angle;

[0077] The positions in the standard frame that do not overlap with the recognized images are filled with solid colors to obtain a standard image.

[0078] Through the above technical solution, the preset picture is placed in the standard frame, and the standard picture formed after filling can be recognized in a unified form, which helps to reduce the complexity of recognition. In addition, the adjustment of the preset angle can make the relevant modules of the recognition process remain unified during training, thereby further reducing the recognition pressure.

[0079] Because the recognition sensitivity of machine equipment to images is completely different from that of the human eye, two images that appear to be completely the same to the human eye may be input at the same time and the output judgments may be completely different. This is because of the difference in the pixel level in the image, because as long as the images are obtained at different times, there must be differences in pixels. In addition, because when training the neural network model used for recognition, the training samples used are obtained in the same way as the standard images, the setting of the preset angle can improve the uniformity of the training samples, thereby reducing the training samples and the amount of training while ensuring recognition accuracy.

[0080] As a further solution of the present invention: a facial feature recognition security control system based on machine learning, comprising:

[0081] A sampling module, used to obtain a sample video including a target face;

[0082] A feature marking module, connected to the sampling module, is used to set a dynamic mark at a specified position of the target face;

[0083] A processing module, which obtains a standard image corresponding to the sampled video according to the dynamic identifier;

[0084] The recognition output module is used to determine the security level of the target face based on the standard image, and adjust and update the specified position according to the security level.

[0085] Among them, the recognition output module is a trained artificial neural network.

[0086] Part types include eyebrows, nose bridge, eyes, and lips.

[0087] The dynamic mark that can be set on the target face of the present invention can be changed according to the security level obtained by subsequent judgment, so as to further re-acquire and identify the standard picture, and take the security level of subsequent judgment as the real judgment result. That is to say, the dynamic mark that can be changed determines that the standard picture will also change accordingly, thereby increasing the probability of changing the second security level judgment. External users cannot accurately know the changing logic of the dynamic mark, and therefore cannot accurately make corresponding disguises, thereby improving the overall security performance.

[0088] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A facial feature recognition security control method based on machine learning, characterized in that: The steps include: S100, obtaining a sample video including a target face; S200, setting a dynamic mark at a designated position of the target face; The step S200 includes: Obtaining a sampling frame in the sampled video; S300, acquiring a standard picture corresponding to the sampled video according to the dynamic identifier; S400, determining the security level of the target face according to the standard image, and adjusting and updating the designated position according to the security level; The step S300 includes: For any of the sampling frames, selecting a preset range encircled by the dynamic mark in the sampling frame as a selected picture; Combining the selected images corresponding to all the sampling frames within a preset time into a recognition image according to a synthesis rule; Processing the identified image into a standard image according to preset rules; The synthesis rules include: Merging the selected images according to the maximum overlap of identical pixels; The same pixels remain unchanged, and the middle value of the different pixels is taken to obtain the recognition image; The preset rules include: Take the standard framework; placing the identification picture in the standard frame at a preset angle; The positions in the standard frame that do not overlap with the identified picture are filled with a solid color to obtain a standard picture.

2. The facial feature recognition security control method based on machine learning according to claim 1 is characterized in that: The sample video includes a facial guidance frame, and the facial guidance frame is used to guide the user to adjust the facial position; When the user's facial position is consistent with the facial guidance frame, activating the dynamic identification; The timestamp of the sampling frame is included in the start time of the dynamic identifier.

3. The facial feature recognition security control method based on machine learning according to claim 2 is characterized in that: The step S200 includes: The designated positions include positions between adjacent part types, and the part types include eyebrows, nose bridge, eyes, and lips.

4. A facial feature recognition security control system based on machine learning, running the facial feature recognition security control method as claimed in any one of claims 1 to 3, characterized in that: The facial feature recognition security control system comprises: A sampling module, used to obtain a sample video including a target face; A feature marking module, connected to the sampling module, for setting a dynamic mark at a designated position of the target face; A processing module, which obtains a standard picture corresponding to the sampled video according to the dynamic identifier; The recognition output module is used to determine the security level of the target face according to the standard image, and adjust and update the designated position according to the security level. Wherein, the recognition output module is a trained artificial neural network.

Images