WO2009003314A1 - Method for realizing personal face login system - Google Patents

Abstract

A method for realizing personal face login system, it carries legible process beforehand using Retinex algorithm for the obtained images in the registering procedure, detects and positions the personal face using a gray-level format image data, then saves the feature vectors of personal face in six difference angles as the feature vectors to be compared with the feature vectors of user's personal face which obtained in the login procedure, determines whether or not allow the user to entry the system basis on the compared result.

Description

 Method for implementing a face login system

 Technical field

 The present invention relates to a login method, and more particularly to a method for implementing a face login system (e.g., an operating system, a game system, etc.).

 Background technique

 At present, the most commonly used method for logging in various systems (such as Windows Unix and other operating systems, game systems, etc.) is to input a password through the keyboard, but this method is not high in security and flexible, and is easily stolen. And easy to forget.

 In order to improve security, a method of using a fingerprint to log in to the system has been proposed, but this method has the following disadvantages: that is, since the fingerprint of the person is thin and the density is high, the collection of the fingerprint data is difficult, and when the user logs in, When there is sweat or dirt on the finger, it will have a great influence on the recognition result, which causes great inconvenience in use. Therefore, it is not easy to obtain such a registration method, and it is not intuitive enough to be applied to the actual application.

 Although some people have proposed the method of applying face to log in to various systems, they often do not give specific implementation methods, and their rejection rate and misrecognition rate are very high, almost 10% or more, so Applicability.

 Summary of the invention

 The technical problem to be solved by the present invention is to provide a method for implementing a face login system, which can enhance the security of system login, and can ensure the applicability, convenience and reliability of login.

 In order to solve the above technical problem, the present invention provides a method for implementing a face registration system, including: a face registration process and a face registration process, where the face registration process includes:

(1) using the Retinex algorithm to process each frame of image data acquired during registration Step

 (2) a step of converting the image data processed by Retinex into a grayscale format;

 (3) Obtaining the face of the face, the angle of the left turn of the face is in the range of 5 to 12 degrees, the angle of the right turn of the face is in the range of 5 to 12 degrees, the angle of the face raising is in the range of 5 to 10 degrees, and the angle of the face is lowered. Steps of feature vectors T1, Τ2, Τ3, Τ4, Τ5, and Τ6 of six face images in the range of 5 to 10 degrees; the face registration process includes:

 (a) using the Retinex algorithm to process each frame of image data acquired at login

(b) the step of converting the Retinex processed image data into a grayscale format;

 (c) Find the front of the face, the angle of the left turn of the face is in the range of 5 to 12 degrees, the angle of the right turn of the face is in the range of 5 to 12 degrees, and the angle of the face is raised in the range of 5 to: L0 or the face a step of lowering an angle of a face image in the range of 5 to 10 degrees;

 (d) a step of acquiring the feature vector Te of the found face image;

 (e) comparing the feature vector Te with the feature vector T1, Τ2, Τ3, Τ4, Τ5 or Τ6 to determine whether the system can be successfully logged into the system.

The present invention adopts the above technical solution, and has the following technical effects, that is, the Retinex method is used to perform clearing processing on the acquired image in advance in the registration process, and the image data in the gray format is used for face detection and positioning, Then, the feature vector of the face at six different angles is saved as a compared feature vector to compare with the user face feature vector obtained during the login process to determine whether the user is allowed to enter the system; The applicability and reliability of the face registration system enable the false recognition rate of face registration to reach one thousandth and the rejection rate to reach one thousandth. Therefore, the system login is ensured and improved. Safety Sexuality, intuitiveness and convenience; In addition, the method of the invention is relatively simple and easy to generalize.

 DRAWINGS

 The present invention will be further described in detail with reference to the accompanying drawings and specific embodiments. FIG. 1 is a schematic diagram of a flow of face registration according to an embodiment of the present invention; FIG. 2 is a diagram of face registration according to an embodiment of the present invention. Schematic diagram of the process.

 detailed description

 Preferably, in order to be able to ensure the accuracy of the face registration system, before using the face to implement system login, the process of face registration in the system is first completed.

 As shown in FIG. 1, in a preferred embodiment, the process of face registration in various systems (such as Windows, Unix, etc., game systems, etc.) is as follows:

 Used to start the camera and display the video first after opening the face registration program in the system. Then, each frame of image data in the video is acquired, so that in the case of poor lighting conditions, the user of the login system can also successfully register according to the present invention and accurately pass the login authentication, preferably using the Retinex algorithm. Each frame of image data is subjected to image enhancement preprocessing to highlight dark areas in the acquired image, improve the visual effect of the image, enhance the contrast of the image, sharpen the edges of the image, and remove noise in the image, thereby The image is clearer and more suitable for subsequent recognition processing.

 Subsequently, the brightness of the image processed by the Retinex algorithm is converted to a range of brightness suitable for face recognition, for example, in one embodiment, the image gradation can be adjusted to be in the range of 120 to 210.

Since the grayscale processing of the image can remove the influence of different illumination on the face image to some extent, in order to ensure that the subsequent face detection can be performed normally, the image data is converted into grayscale. Format. In one embodiment, the conversion of the grayscale format can be implemented using the following formula:

Y = k _r R + k _g G + k _b B

Where R, G, B represent three components of the color image; k _r , k _g , k _b respectively represent the weights corresponding to the respective components; Y is the converted gray image. In one embodiment, k _r =0. 299, k _g =0.587, k _b =0. 114.

 Based on the obtained grayscale map data, face detection is performed to determine data in which the face is represented. In one embodiment, the Adaboost algorithm can be used for face detection. Then, in the data determined as the face, the position of the organ in the face is located, optimally, in the present invention, the eye is positioned; in one embodiment, the Adaboost algorithm can also be used to implement the organ in the face. Positioning. The Adaboost algorithm has the characteristics of simple method, good real-time performance and fast detection speed. It is a classifier algorithm. The basic implementation idea is: Use a large number of general classifiers with basic classification ability to superimpose through certain methods ( Boost), constitute a strong classifier with strong classification ability, and then connect several strong classifiers into a classifier cascade to complete image search detection. Therefore, Adaboost can be used to ensure face detection and people. The accuracy of face positioning can also improve the speed of face detection and face positioning. Since it has been found in experiments that the selection of system detection parameters has a great influence on the detection speed and the recognition effect, in one embodiment, the extended Hair-like feature can be extracted to complete the detection and localization of the face.

Then, determining the position of the face image data of each frame subjected to the above processing, that is, calculating the angle at which the face is deflected, and preferably, selecting the six representative angle images of the face, respectively, which are respectively Yes: A photo of the front of the face, an image with a left-angle of the face in the range of 5 to 12 degrees, an image with a right-angle of the face in the range of 5 to 12 degrees, and a face-lifting angle An image in the range of 5 to 10 degrees and a picture of the face with a face angle of 5 to 10 degrees. Since the user has a high probability that the orientation angle of the face appears within the above-mentioned range of angles during the login process, it is helpful to improve the user's authentication during the login system by selecting the six face images as described above. speed.

Then, the face data in the above six face images are respectively taken out and subjected to illumination compensation processing. In one embodiment, the face data may be subjected to illumination compensation processing using the following formula: _S = T (r) = l / [l + (m + _r ) ^E ], where r represents the brightness of the input image, and s is the output image For the corresponding brightness value, E controls the slope of the function, and m is the gray level.

 The data of the six images after illumination compensation are respectively Gaubor transformed, and then projected to PCA (principal component analysis) to project the samples of the Gabor transformed high-dimensional space into the low-dimensional space; then transform to LDA (linear discrimination) Analyze) space to extract the most discriminative low-dimensional features, and finally form the facial feature values into facial feature vectors, and extract the feature vectors of the above six images: front Tl, left turn 5-12 degrees Τ 2, turn right 5-12 degrees Τ 3, head up 5-10 degrees Τ 4, head 5-10 degrees Τ 5, left or right 5-10 degrees Τ 6.

In one embodiment, the present invention selects the Gabor wavelet transform most commonly used by researchers in the field of computational vision, the kernel function of which is described as:

Among them, "representing the scale parameter and direction parameter of Gabor convolution kernel respectively; z = ; , _ν = ^ ^Φ ', where corpse, A _max is the maximum frequency; / is the spatial factor of Gabor kernel in the frequency domain.

Finally, the obtained six face images and their corresponding feature vectors T1, Τ2, Τ3, Τ4, T5 and T6 are saved in file or data mode, and one of three usage modes: "loose", "standard", and "strict" is selected for saving. In one embodiment, the above three modes of use respectively correspond to "85". The three similarities of ", "90" and "95" can be selected by the user according to the requirements of the system confidentiality.

 After the face registration is completed in the system, as shown in FIG. 2, in a preferred embodiment, the user can log in to the system by the following method:

 When the system is turned on, the face login system program is started at the same time, and then the camera and the display are activated, and then the image data of each frame in the video is acquired. In order to make the user of the login system accurately pass under the condition of poor lighting conditions, Login authentication, preferably, using the Retinex algorithm to perform image enhancement preprocessing on each frame of image data acquired to highlight dark areas in the acquired image, improve the visual effect of the image, enhance the contrast of the image, and make the edges of the image clear And removing the noise in the image, so that the converted image is more clear, and is more suitable for subsequent recognition processing.

 Subsequently, the brightness of the Retinex processed image is converted to a range of brightness suitable for face recognition, for example, in one embodiment, the image gradation can be adjusted to a range of 120 to 210.

 Since the gradation processing of the image can remove the influence of different illumination on the face image to some extent, in order to ensure that the subsequent face detection can be performed normally, the image data is converted into a grayscale format. In one embodiment, the following formula can be used to implement the conversion of the grayscale format -

Y two k _r R + k _g G + k _b B The weight corresponding to the component; Y is the converted grayscale image. In one embodiment, k 0. 299, k 0., 587, k _b =0. 114 may be taken.

 Based on the obtained grayscale format image data, face detection is performed to determine data in which the face is represented. In one embodiment, the Adaboost algorithm can be used for face detection. Then, in the data determined as the face, the position of the same organ in the face as the registered organ is located, and optimally, in the present invention, the eye is selected accordingly; in one embodiment, The Adaboost algorithm can be used to locate the organs in the face. Since it has been found through experiments that the selection of system detection parameters has a great influence on the detection speed and recognition effect, in one embodiment, the extended Hair-like feature can be extracted to complete the detection and localization of the face.

 Then, determining the position of the image data subjected to the above processing, that is, calculating the angle at which the face is deflected, and judging from the real-time whether the image data is a face in a range of six angles as described below: that is, whether it is a face The image on the front side, the image in which the left turn angle of the face is in the range of 5 to 12 degrees, the image in which the face is rotated right in the range of 5 to 12 degrees, the image in which the face is raised in the range of 5 to 10 degrees or the face is An image with a head angle in the range of 5 to 10 degrees. As long as you find a face image within the above angle range, you can proceed to the next step.

 The face image found in the previous step is taken, and the face data is taken out according to the position of the face of the face, and the data is subjected to illumination compensation processing.

 Perform Gabor transform on the data after illumination compensation, and then project to PCA (principal component analysis) to project the samples of Gabor transformed high-dimensional space into low-dimensional space; then transform into LM (linear discriminant analysis) space. The face feature vector Te is formed.

In one embodiment, the present invention selects the Gabor wavelet transform most commonly used by researchers in the field of computational vision, the kernel function of which is described as:

Where, "represents the scale parameter and direction parameter of the Gabor convolution kernel respectively; _z = (x, j;) _; , _v = k _v e ⁱ , where =^ / corpse, _ν = Μ /8, is the maximum frequency; / is the spatial factor of the Gabor kernel in the frequency domain.

 The face feature vector Te in the video image data is taken out, and Te is compared with the St-symbol vectors T1, Τ2, Τ3, Τ4, Τ5, and Τ6 stored at the time of registration, and is used as a threshold according to the usage mode selected by the face registration. Determine if the identification is successful. The recognition threshold is determined based on the save mode selected in the face registration process as a threshold.

 Once the identification is successful, you can automatically log in to the operating system. If the recognition fails, the face Pe that identifies the failure is saved (for querying the face whose display recognition is failed), and the Retinex processing and the following steps are repeated for each frame image acquired from the video to perform real-time recognition.

Claims

Rights request  A method for implementing a face login system, comprising: a face registration process and a face login process, wherein:  The face registration process includes:  (1) a step of processing each frame of video image data acquired at the time of registration using the Retinex algorithm;  (2) a step of converting the image data processed by Retinex into a grayscale format;  (3) Obtaining the face of the face, the angle of the left turn of the face is in the range of 5 to 12 degrees, the angle of the right turn of the face is in the range of 5 to 12 degrees, the angle of the face raising is in the range of 5 to 10 degrees, and the angle of the face is lowered. Steps of feature vectors T1, Τ2, Τ3, Τ4, Τ5, and Τ6 of six face images in the range of 5 to 10 degrees; the face registration process includes:  (a) a step of processing each frame of image data acquired at the time of login using the Retinex algorithm;  (b) the step of converting the Retinex processed image data into a grayscale format;  (c) Find the front of the face, the angle of the left turn of the face is in the range of 5 to 12 degrees, the angle of the right turn of the face is in the range of 5 to 12 degrees, the angle of the face is raised in the range of 5 to 10 degrees, or the face is bowed. a step of an image of a face having an angle in the range of 5 to 10 degrees;  (d) a step of acquiring the feature vector Te of the found face image;  (e) comparing the feature vector Te with the feature vector T1, Τ2, Τ3, Τ4, Τ5 or Τ6 to determine whether the system can be successfully logged into the system. The method for implementing a face registration system according to claim 1, further comprising: between the step (1) and the step (2): the image processed by the step (1) of The step of converting the brightness into a range of brightness suitable for face recognition;  Between the step (a) and the step (b), the method further comprises: converting the brightness of the image processed in the step (a) to a range of brightness suitable for face recognition.  The method for implementing a face registration system according to claim 1 or 2, wherein the step (3) further comprises:  (i) performing steps of face detection and face positioning according to the grayscale format image data obtained in step (2);  (ii) determining the position of the face image data of each frame, that is, calculating the angle in which the face is deflected, and selecting six face images, the six face images are: face front, face a face with a left turn angle in the range of 5 to 12 degrees, a face right turn angle in the range of 5 to 12 degrees, a face face lift angle in the range of 5 to 10 degrees, and a face face angle in the range of 5 to 10 degrees One image each;  (iii) taking out the face data in the six face images separately and performing the step of performing illumination compensation processing;  (iv) Gabor transform is performed on the face data obtained in step (iii), and then projected into the principal component analysis space and the linear discriminant analysis space to obtain a face feature vector composed of facial feature values, and then the corresponding face is extracted therefrom. Feature vectors T1, Τ2, Τ3, Τ4, Τ5, and Τ6.  The method for implementing a face registration system according to claim 3, wherein in the step (i), the Adaboost algorithm is used to implement face detection and face localization. The method for implementing a face registration system according to claim 1 or 4, wherein the face registration process further comprises: using the obtained six face images and their corresponding face feature vectors Tl, Τ2, Τ3, Τ4, Τ5, and Τ6 are saved as files or data. The method for implementing a face registration system according to claim 5, wherein the similarity is used when saving the face image and its feature vectors T1, Τ2, Τ3, Τ4, Τ5, and Τ6 Save for one of "85", "90" or "95".  The method for implementing a face login system according to claim 1 or 2, wherein the step (c) comprises:  Performing steps of face detection and face localization according to the grayscale format image data obtained in step (b);  Determine the position of the face image data of each frame, that is, calculate the angle at which the face is deflected, and find out that it is included in the front of the face, the left turn angle of the face is in the range of 5 to 12 degrees, and the right turn angle of the face is The step of a face image data in the range of 5 to 12 degrees, the face heading angle is in the range of 5 to: L0 degrees or the face head angle is in the range of 5 to 10 degrees.  The method for implementing a face registration system according to claim 7, wherein the face detection and the face location are implemented using an Adaboost algorithm.  The method for implementing a face login system according to claim 1 or 2, wherein the step (d) comprises:  Extracting the face data of the face image found in step (c), and performing the step of performing illumination compensation processing on the data;  The obtained face data is Gabor transformed, and then projected into the principal component analysis space and the linear discrimination analysis space to form a corresponding face feature vector Te.  10. The method for implementing a face registration system according to claim 6, wherein in the step (e), the storage mode selected in the face registration process is used as a threshold. Determine if you can successfully log in to the system.