Face recognition system based on deep learning

Abstract

The invention discloses a face recognition system based on deep learning. The face recognition system comprises a first face acquisition module, a second face acquisition module, a third face acquisition module, a data processing module and an information sending module, the first face acquisition module is used for acquiring first face information at a first preset position, the second face acquisition module is used for acquiring second face information at a second preset position, and the third face acquisition module is used for acquiring third face information at a third preset position; and the data processing module is used for processing the first face information, the second face information, the third face information and the equipment information to obtain first identification information, second identification information, third identification information, overall identification information and equipment warning information. According to the invention, face recognition can be better carried out, and different use requirements of users are met.

Description

Technical Field

[0001] The present invention relates to the field of database security, and in particular to a face recognition system based on deep learning. Background Art

[0002] Face recognition is a biometric technology that identifies people based on their facial features. It uses a camera or camcorder to collect images or video streams containing faces, automatically detects and tracks faces in images, and then performs facial recognition on the detected faces. It is also commonly called portrait recognition or facial recognition.

[0003] During the face recognition process, the face recognition system will be used.

[0004] The existing face recognition system has a single recognition method and cannot meet the actual usage needs, which has a certain impact on the use of the face recognition system. Therefore, a face recognition system based on deep learning is proposed. Summary of the invention

[0005] The technical problem to be solved by the present invention is: how to solve the problem that the existing face recognition system has a single recognition method and cannot meet the actual use requirements, which has a certain impact on the use of the face recognition system, and provides a face recognition system based on deep learning.

[0006] The present invention solves the above technical problems through the following technical solutions. The present invention includes a first face acquisition module, a second face acquisition module, a third face acquisition module, a data processing module and an information sending module;

[0007] The first face acquisition module is used to acquire first face information at a first preset position, the second face acquisition module is used to acquire second face information at a second preset position, and the third face acquisition module is used to acquire third face information at a third preset position;

[0008] The data processing module is used to process the first face information, the second face information, the third face information and the device information to obtain the first identification information, the second identification information, the third identification information, the overall identification information and the device warning information;

[0009] The information sending module is used to send the first identification information, the second identification information, the third identification information, the overall evaluation information and the equipment warning information to a preset receiving terminal after the above information is generated.

[0010] Furthermore, the specific processing process of the first identification information is as follows: extract the collected first facial information, the first facial information is the image information of the left side of the face, process the image information of the left side of the face to obtain a first real-time evaluation parameter, compare the first real-time evaluation parameter with the data in the preset data, calculate the difference between the first real-time evaluation parameter and the standard first parameter preset in the preset data, and obtain the first identification information.

[0011] Furthermore, the specific processing process of the second identification information is as follows: extract the collected second facial information, the second facial information is the image information of the right side of the face, process the image information of the right side of the face to obtain a second real-time evaluation parameter, compare the second real-time evaluation parameter with the data in the preset data, calculate the difference between the second real-time evaluation parameter and the standard second parameter preset in the preset data, and obtain the second identification information.

[0012] Furthermore, the specific processing process of the third identification information is as follows: extract the collected third facial information, the third facial information is the frontal face image information of the face, process the frontal face image information to obtain the third real-time evaluation parameter, compare the third real-time evaluation parameter with the data in the preset data, calculate the difference between the third real-time evaluation parameter and the standard third parameter preset in the preset data, and obtain the third identification information.

[0013] Furthermore, the specific processing process of processing the left side image information of the face to obtain the first real-time evaluation parameter is as follows: extracting the collected left side image information of the face, performing a clear processing on the left side image information to obtain clear left side face information, and performing feature point extraction on the clear left side face information;

[0014] Mark the nose tip as point a1, the leftmost point of the left nose wing as point a2, and the left corner of the left eye and the right corner of the left eye as points a3 and a4;

[0015] Connect points a1 and a2 to obtain line segment L1, connect points a2 and a3 to obtain line segment L2, connect points a3 and a4 to obtain line segment L3, connect points a4 and a1 to obtain line segment L4. Line segments L1, L2, L3 and L4 form a first identification area. The area of ​​the first identification area is measured to obtain the first real-time evaluation parameter.

[0016] Furthermore, the specific process of processing the right side image information of the face to obtain the second real-time evaluation parameter is as follows: extracting the collected right side image information of the face, performing a clear processing on the right side image information to obtain clear right side face information, and performing feature point extraction on the clear right side face information;

[0017] Mark the nose tip as point b1, mark the rightmost point of the right nose wing as point b2, and mark the left corner of the right eye and the right corner of the right eye as points b3 and b4;

[0018] Connect points b1 and b2 to obtain line segment P1, connect points b2 and b3 to obtain line segment P2, connect points b3 and b4 to obtain line segment P3, connect points b4 and b1 to obtain line segment P4, line segments P1, P2, P3 and P4 form a second recognition area, and the area of ​​the second recognition area is measured to obtain the second real-time evaluation parameter.

[0019] Furthermore, the specific process of processing the frontal face image information to obtain the third real-time evaluation parameter is as follows: extract the collected frontal face image information, perform clarity processing on it to obtain clear face information, perform feature point extraction on the clear face information, mark the outer corners of the eyes on both sides as points c1 and c2, mark the nose tip as c3, connect point c1 and point c2 to obtain line segment M1, connect point c1 and point c2 with point c3 respectively to obtain line segments M2 and M3, line segments M1, line segments M2 and M3 surround a third recognition area, and measure the area of ​​the third recognition area to obtain the third real-time evaluation parameter.

[0020] Furthermore, the overall identification information includes identification success information and identification failure information, and the acquisition process of the overall identification success and the overall identification failure is as follows: extracting the collected first identification information, the second identification information and the third identification information, and when any one of the first identification information and the second identification information is obtained to be less than a preset value, the third identification information is verified, and when the third identification information is also less than the preset value at this time, the identification success information is generated, and when the third identification information is greater than the preset value at this time, the identification failure information is generated;

[0021] When both the first identification information and the second identification information are greater than the preset value, identification failure information is directly generated.

[0022] Furthermore, the device warning information includes first warning information, second warning information and third warning information, and the specific processing process of the device warning information is as follows: extracting the collected first face information, collecting the first face information x times continuously, and then analyzing the clarity information thereof, obtaining the clarity of the x first face information, and when the clarity of the x first face information is less than a preset value and greater than x / 3, the first warning information is generated;

[0023] Extract the collected second face information, collect the first face information x times continuously, then analyze its clarity information, obtain the clarity of x third face information, and when the clarity of the x second face information is less than the preset value and greater than x / 3, generate the second warning information;

[0024] The collected third face information is extracted, the third face information is collected continuously x times, and then the clarity information is analyzed to obtain the clarity of x pieces of third face information. When the clarity of the x pieces of third face information is less than the preset value and greater than x / 3, the third warning information is generated.

[0025] Compared with the prior art, the present invention has the following advantages: the face recognition system based on deep learning can obtain more accurate left face related parameter information through the generated first recognition information, the second recognition information can obtain more accurate right face related parameter information, and the third recognition information can obtain more accurate front face related parameter information, thereby realizing the accurate acquisition of comprehensive face recognition related parameters, ensuring that subsequent face recognition has higher accuracy, and through the generated overall recognition information, accurate face recognition can be performed to reduce the occurrence of recognition errors. At the same time, multi-directional face information collection and recognition settings realize multi-mode faster and more accurate face recognition, and through the generated device warning information, problems with the face image acquisition device can be discovered in time, thereby ensuring that the acquired face image is clearer and reducing the occurrence of face recognition failures due to abnormalities in the acquisition device, making the system more worthy of popularization and use. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a system block diagram of the present invention. DETAILED DESCRIPTION

[0027] The following is a detailed description of an embodiment of the present invention. This embodiment is implemented on the premise of the technical solution of the present invention, and a detailed implementation method and a specific operation process are given, but the protection scope of the present invention is not limited to the following embodiment.

[0028] like Figure 1 As shown, this embodiment provides a technical solution: a face recognition system based on deep learning, including a first face acquisition module, a second face acquisition module, a third face acquisition module, a data processing module and an information sending module;

[0029] The first face acquisition module is used to acquire first face information at a first preset position, the second face acquisition module is used to acquire second face information at a second preset position, and the third face acquisition module is used to acquire third face information at a third preset position;

[0030] The data processing module is used to process the first face information, the second face information, the third face information and the device information to obtain the first identification information, the second identification information, the third identification information, the overall identification information and the device warning information;

[0031] The information sending module is used to send the first identification information, the second identification information, the third identification information, the overall evaluation information and the equipment warning information to a preset receiving terminal after the above information is generated.

[0032] The present invention can obtain more accurate left face related parameter information through the generated first identification information, can obtain more accurate right face related parameter information through the second identification information, and can obtain more accurate front face related parameter information through the third identification information, thereby realizing the accurate acquisition of comprehensive face recognition related parameters, ensuring that subsequent face recognition has higher accuracy, and through the generated overall identification information, accurate face recognition can be performed to reduce the occurrence of recognition errors, and through the generated device warning information, problems with the face image acquisition device can be discovered in time, thereby ensuring that the acquired face image is clearer and reducing the occurrence of face recognition failures due to abnormalities in the acquisition device, making the system more worthy of popularization and use.

[0033] The specific processing process of the first identification information is as follows: extracting the collected first face information, the first face information is the image information of the left side of the face, processing the image information of the left side of the face to obtain a first real-time evaluation parameter, comparing the first real-time evaluation parameter with the data in the preset data, calculating the difference between the first real-time evaluation parameter and the standard first parameter preset in the preset data, that is, obtaining the first identification information;

[0034] Through the above process, more accurate first identification information can be obtained to provide better data support for the final identification.

[0035] The specific processing process of the second identification information is as follows: extracting the collected second face information, the second face information is the image information of the right side of the face, processing the image information of the right side of the face to obtain a second real-time evaluation parameter, comparing the second real-time evaluation parameter with the data in the preset data, calculating the difference between the second real-time evaluation parameter and the standard second parameter preset in the preset data, that is, obtaining the second identification information;

[0036] Through the above process, more accurate second identification information can be obtained to provide better data support for the final identification.

[0037] The specific processing process of the third identification information is as follows: extracting the collected third face information, the third face information is the frontal face image information of the face, processing the frontal face image information to obtain the third real-time evaluation parameter, comparing the third real-time evaluation parameter with the data in the preset data, calculating the difference between the third real-time evaluation parameter and the standard third parameter preset in the preset data, that is, obtaining the third identification information;

[0038] Through the above process, more accurate third-party identification information can be obtained to provide better data support for the final identification.

[0039] The specific processing process of processing the left side image information of the face to obtain the first real-time evaluation parameter is as follows: extracting the collected left side image information of the face, performing a clear processing on the left side image information to obtain clear left side face information, and extracting feature points on the clear left side face information;

[0040] Mark the nose tip as point a1, the leftmost point of the left nose wing as point a2, and the left corner of the left eye and the right corner of the left eye as points a3 and a4;

[0041] Connect points a1 and a2 to obtain line segment L1, connect points a2 and a3 to obtain line segment L2, connect points a3 and a4 to obtain line segment L3, connect points a4 and a1 to obtain line segment L4, line segments L1, L2, L3 and L4 form a first recognition area, and the area of ​​the first recognition area is measured to obtain the first real-time evaluation parameter;

[0042] Through the above process, accurate first evaluation parameters can be acquired, thereby providing data support for the first identification information.

[0043] The specific process of processing the right side image information of the face to obtain the second real-time evaluation parameter is as follows: extracting the collected right side image information of the face, performing a clear processing on the right side image information to obtain clear right side face information, and performing feature point extraction on the clear right side face information;

[0044] Mark the nose tip as point b1, mark the rightmost point of the right nose wing as point b2, and mark the left corner of the right eye and the right corner of the right eye as points b3 and b4;

[0045] Connect points b1 and b2 to obtain line segment P1, connect points b2 and b3 to obtain line segment P2, connect points b3 and b4 to obtain line segment P3, connect points b4 and b1 to obtain line segment P4, line segments P1, P2, P3 and P4 form a second recognition area, and measure the area of ​​the second recognition area to obtain the second real-time evaluation parameter;

[0046] Through the above process, accurate second evaluation parameters can be obtained, thereby providing data support for the second identification information.

[0047] The specific process of processing the frontal face image information to obtain the third real-time evaluation parameter is as follows: extracting the collected frontal face image information, performing a clear processing on it to obtain clear face information, performing feature point extraction on the clear face information, marking the outer corners of the eyes on both sides as points c1 and c2, marking the nose tip as c3, connecting points c1 and c2 to obtain line segment M1, connecting points c1 and c2 to point c3 to obtain line segments M2 and M3, line segments M1, M2 and M3 enclose a third recognition area, and measuring the area of ​​the third recognition area to obtain the third real-time evaluation parameter;

[0048] Through the above process, accurate third evaluation parameters can be obtained, thereby providing data support for the third identification information.

[0049] The sharpening process in the above process is an image sharpening process based on deep learning, and the processing process is as follows: training data. In order to train the model, a large amount of high-definition videos and low-definition videos need to be prepared as training data, and the model can learn the mapping relationship from low-definition to high-definition through comparative learning.

[0050] Model selection and training. Commonly used models include self-attention models, convolutional neural networks (CNNs), and recurrent neural networks (RNNs). These models can model the temporal and spatial characteristics of videos, thereby better restoring the clarity of videos.

[0051] Loss function design. In order to constrain the training of the model, it is necessary to design a suitable loss function, such as the mean square error (MSE) or structural similarity (SSIM) at the pixel level.

[0052] Model optimization: Through the back-propagation algorithm and optimizer, the model parameters are continuously adjusted to minimize the loss function, thus obtaining an optimized model.

[0053] Application and testing. Applying the optimized model to low-definition videos can significantly improve the video clarity. At the same time, the effectiveness of the model can also be verified through comparative experiments.

[0054] The overall identification information includes identification success information and identification failure information. The acquisition process of the overall identification success and the overall identification failure is as follows: extract the collected first identification information, the second identification information and the third identification information. When any one of the first identification information and the second identification information is less than a preset value, the third identification information is verified. When the third identification information is also less than the preset value, identification success information is generated. When the third identification information is less than the preset value, identification failure information is generated.

[0055] When both the first identification information and the second identification information are greater than the preset value, identification failure information is directly generated.

[0056] The device warning information includes first warning information, second warning information and third warning information. The specific processing process of the device warning information is as follows: extracting the collected first face information, collecting the first face information x times continuously, and then analyzing its clarity information to obtain the clarity of the x first face information. When the clarity of the x first face information is less than the preset value and greater than x / 3, the first warning information is generated;

[0057] Extract the collected second face information, collect the first face information x times continuously, then analyze its clarity information, obtain the clarity of x third face information, and when the clarity of the x second face information is less than the preset value and greater than x / 3, generate the second warning information;

[0058] Extract the collected third person's face information, collect the third person's face information x times continuously, then analyze the clarity information, obtain the clarity of x pieces of third person's face information, and when the clarity of the x pieces of third person's face information is less than a preset value and greater than x / 3, generate the third warning information;

[0059] Through the above process, it is possible to timely discover whether there is an abnormality in the facial image acquisition device, thereby reducing facial recognition errors caused by abnormalities in the facial acquisition device.

[0060] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0061] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0062] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A face recognition system based on deep learning, characterized by: It includes a first face acquisition module, a second face acquisition module, a third face acquisition module, a data processing module and an information sending module; The first face acquisition module is used to acquire first face information at a first preset position, the second face acquisition module is used to acquire second face information at a second preset position, and the third face acquisition module is used to acquire third face information at a third preset position; The data processing module is used to process the first face information, the second face information, the third face information and the device information to obtain the first identification information, the second identification information, the third identification information, the overall identification information and the device warning information; The information sending module is used to send the first identification information, the second identification information, the third identification information, the overall evaluation information and the equipment warning information to a preset receiving terminal after the above information is generated.

2. The face recognition system based on deep learning according to claim 1, characterized in that: The specific processing process of the first identification information is as follows: extract the collected first facial information, the first facial information is the image information of the left side of the face, process the image information of the left side of the face to obtain a first real-time evaluation parameter, compare the first real-time evaluation parameter with the data in the preset data, calculate the difference between the first real-time evaluation parameter and the standard first parameter preset in the preset data, and obtain the first identification information.

3. The face recognition system based on deep learning according to claim 1, characterized in that: The specific processing process of the second identification information is as follows: extract the collected second facial information, the second facial information is the image information of the right side of the face, process the image information of the right side of the face to obtain a second real-time evaluation parameter, compare the second real-time evaluation parameter with the data in the preset data, calculate the difference between the second real-time evaluation parameter and the standard second parameter preset in the preset data, and obtain the second identification information.

4. The face recognition system based on deep learning according to claim 1, characterized in that: The specific processing process of the third identification information is as follows: extract the collected third facial information, the third facial information is the frontal face image information of the face, process the frontal face image information to obtain the third real-time evaluation parameter, compare the third real-time evaluation parameter with the data in the preset data, calculate the difference between the third real-time evaluation parameter and the preset standard third parameter in the preset data, and obtain the third identification information.

5. The face recognition system based on deep learning according to claim 2, characterized in that: The specific processing process of processing the left side image information of the face to obtain the first real-time evaluation parameter is as follows: extracting the collected left side image information of the face, performing a clear processing on the left side image information to obtain clear left side face information, and extracting feature points on the clear left side face information; Mark the nose tip as point a1, the leftmost point of the left nose wing as point a2, and the left corner of the left eye and the right corner of the left eye as points a3 and a4; Connect points a1 and a2 to obtain line segment L1, connect points a2 and a3 to obtain line segment L2, connect points a3 and a4 to obtain line segment L3, connect points a4 and a1 to obtain line segment L4. Line segments L1, L2, L3 and L4 form a first identification area. The area of ​​the first identification area is measured to obtain the first real-time evaluation parameter.

6. The face recognition system based on deep learning according to claim 3, characterized in that: The specific process of processing the right side image information of the face to obtain the second real-time evaluation parameter is as follows: extracting the collected right side image information of the face, performing a clear processing on the right side image information to obtain clear right side face information, and performing feature point extraction on the clear right side face information; Mark the nose tip as point b1, mark the rightmost point of the right nose wing as point b2, and mark the left corner of the right eye and the right corner of the right eye as points b3 and b4; Connect points b1 and b2 to obtain line segment P1, connect points b2 and b3 to obtain line segment P2, connect points b3 and b4 to obtain line segment P3, connect points b4 and b1 to obtain line segment P4, line segments P1, P2, P3 and P4 form a second recognition area, and the area of ​​the second recognition area is measured to obtain the second real-time evaluation parameter.

7. The face recognition system based on deep learning according to claim 1, characterized in that: The specific process of processing the frontal face image information to obtain the third real-time evaluation parameter is as follows: extract the collected frontal face image information, perform clarity processing on it to obtain clear face information, perform feature point extraction on the clear face information, mark the outer corners of the eyes on both sides as points c1 and c2, mark the tip of the nose as c3, connect points c1 and c2 to obtain line segment M1, connect points c1 and c2 with point c3 respectively to obtain line segments M2 and M3, line segments M1, M2 and M3 surround a third recognition area, and measure the area of ​​the third recognition area to obtain the third real-time evaluation parameter.

8. The face recognition system based on deep learning according to claim 1, characterized in that: The overall identification information includes identification success information and identification failure information. The acquisition process of the overall identification success and the overall identification failure is as follows: extract the collected first identification information, the second identification information and the third identification information. When any one of the first identification information and the second identification information is less than a preset value, the third identification information is verified. When the third identification information is also less than the preset value, identification success information is generated. When the third identification information is less than the preset value, identification failure information is generated. When both the first identification information and the second identification information are greater than the preset value, identification failure information is directly generated.

9. The face recognition system based on deep learning according to claim 1, characterized in that: The device warning information includes first warning information, second warning information and third warning information. The specific processing process of the device warning information is as follows: extracting the collected first face information, collecting the first face information x times continuously, and then analyzing its clarity information to obtain the clarity of the x first face information. When the clarity of the x first face information is less than the preset value and greater than x / 3, the first warning information is generated; Extract the collected second face information, collect the first face information x times continuously, then analyze its clarity information, obtain the clarity of x third face information, and when the clarity of the x second face information is less than the preset value and greater than x / 3, generate the second warning information; The collected third face information is extracted, the third face information is collected continuously x times, and then the clarity information is analyzed to obtain the clarity of x pieces of third face information. When the clarity of the x pieces of third face information is less than the preset value and greater than x / 3, the third warning information is generated.

Images