The Research on Face Recognition and Segmentation Based on Intelligent Background

Affected by factors such as attitude, light, expression, etc., it is impossible to accurately identify the identity in a wireless visual sensor network in an uncontrollable environment. In traditional visual identity recognition, it is necessary to convert uncontrollable factors into controllable and stable feature factors for identity recognition in a relatively uncontrollable environment where the node distribution is relatively complicated. The conversion process leads to long recognition time and low efficiency. An adaptive recognition method for identity features in wireless visual sensing networks based on LBP face recognition is proposed. A strong classifier is obtained for cascading, and the underlying features are extracted. The final Harr face cascade classifier is applied to the face Check it out. The PCA dimensionality reduction processing of the facial area feature vector is performed to obtain the low-dimensional feature vector, the dimensionality reduction coefficient, and the average face of the person. For the face image in the wireless local area, its LBP operation is given. Perform histogram statistics on face feature information, obtain face LBP histograms, and perform feature matching on the face feature database to complete recognition. The improved algorithm has improved the cumulative matching score of traditional algorithms by 17.8%; the accuracy rate has improved It is 32.7%, and the recognition time is shortened by 3.9s. Simulation results show that the proposed algorithm has high accuracy and recognition efficiency.


Introduction
As wireless visual sensor network development and the application of information security people put forward higher requirements [1]. Licensing is one of the important means to ensure information security, it may be advantageous to protect users of the system have the corresponding authority [2,3]. Identity adaptive knowledge is the key to the theoretical basis of authorization, recognition algorithm research topics have become the focus of scholars in the field, it has been more and more attention [4,5].
Identity recognition technology as a key development issues of network security, common identity recognition technology including face recognition, iris recognition, fingerprint recognition. Correlation algorithm achieved certain results. Among them, the literature [6] Proposed single feature identity recognition algorithm, based on a certain facial features to identify the identity of the algorithm has fast calculation speed, real-time advantage, but the affected light, rotation, etc., in a complex context, recognition rate will be reduced; literature [7] Proposed identity recognition algorithm combined features will not be fixed in accordance with the same characteristics as the principles together, and this algorithm is more efficient overall, but the information is too large, resulting in reduced real-time recognition; literature [8] Proposed common vector combined identity 2 DPCA special algorithm, through Uram-Schmidt orthogonal transformation implemented on the same vector image is calculated, based on the results obtained to identify the minimum distance detection, the algorithm has a good discrimination function, but complexity is too high, low recognition efficiency; literature [9] Proposed a three-dimensional model of facial expression recognition algorithm identity of deformed face implementation model registration, high accuracy, there are common face model and real-time problem of poor matching model registration it can not be obtained; literature [10] SFM's proposed identity recognition algorithm, the first search match point, after SFM algorithm determined the three-dimensional coordinates of the feature points, using a multi-resolution image mosaic texture image forming algorithm, while the maps them to three-dimensional model, the algorithm increases the real sense, but a three-dimensional reconstruction of a large amount of computation, greatly affected the performance of its recognition.
Proposed LBP face discrimination in wireless visual sensor network identity characteristic adaptive algorithm to identify, by Harr face cascade classifier to detect human faces, human face region feature vector PCA dimension reduction, for wireless local area face images, use LBP get facial feature information obtained facial feature information histogram statistics, get face LBP histogram, which will feature visual and wireless sensor networks signatures were human face match, to complete the identification of the identity. Experimental results demonstrate that the proposed algorithm has high recognition accuracy and efficiency.

Visual Identity Adaptive Wireless Sensor Networks Recognition Principle
When identity adaptive recognition through the inspection window in a customized location adjacent rectangular matrix of each pixel is obtained, at the same time seeking the difference, and then after the implementation of the recognition of the difference between the obtained sub-areas of the image.
In the identification process focused contain edge characteristics, central feature of the linear characteristic three samples, The total of 14 fundamental characteristic of the prototype.
If the detection window size W H × , With quad rectangular description: ( ) recSum r is used to describe graphics window and all pixels, the rectangle feature can be described as follows: Because the process of feature recognition calculation, the statistics need all the pixel values in a rectangle and makes training and testing efficiency is greatly reduced, therefore, we need to integrate special parameters to reduce property value calculating expenses, increase efficiency. Highlights include integral image angle is 0° and 45° the two ramps FIG. 0° inclination is used below figure shows an example of integration, Transmission ( ) recSum r evolution. For angle of 0° is a characteristic square, integral image ( ) , SAT x y can be described as: , , , I x y is used to describe the pixels value of just press rows or columns to obtain a travers: For feature matrix: , through all of the pixels of the matrix and can be obtained target identification equation: The analysis equation, matrix r 's ( ) recSum r . Only time will be able to obtain integral image search, computational efficiency, while high recognition rate.

Face Cascade Classifier
Harr Where in, m and l were used to describe the number of the human face of the sample， n m l = + ; 3) By the following formula the weights are normalized: 5) Select the minimum error rate of t ε , classer t h ; 6) The weights are updated: 7) The final strong classifier can be described as: Harr face cascade classifier strong classifier through the implementation of the sub-window to select image input face image after image fusion under the sub-window level in continued implementation of strong classifier selection, Finally, obtained with a high accuracy rate of face sub-window image.

Face Image Processing PCA Dimension Reduction
Harr face cascade respectively acquires face image face range message contains a plurality of Harr-like features, you want to increase the rear face alignment efficiency shall conclude the face range characteristic vectors. PCA dimension reduction governance, to obtain a low-dimensional feature vector and maintain lower number. And then the man's face in the field after the dimensionality reduction characteristic vector is normalized control, get the person's average face. PCA dimension reduction method of face range message is as follows: 1) First, the face image of the sample by column expands into a one-dimensional vector, the calculation of their average image vector: There in， i x is used to describe a face samples Vector; M is used to describe the total sample; m Used to describe the average vector image, in the face image of the sample it is commonly referred to as the average face.
2) Covariance matrix of training samples is calculated: For ease of calculation, the paper by the singular value decomposition theorems obtained eigenvalues, Where, i α is used to describe the projection coefficients corresponding to feature vectors, which can reflect the input image feature amount of the position. Then for any image, rectangular Feature 1 reduce the dimension after dimension reduction feature can be described as:

Algorithm Implement
LBP operator is a useful texture description operator, which uses the gray value associated with the size of the face image of each pixel with its neighbor points consisting of the binary-coded image of local texture space formed to express, not having a rotation denaturation and brightness invariant. LBP operator can be described as follows: The name is a key point, then the size of the 3 × 3 window is called the gray this critical point is called the gray value of the eight-point threshold, and the associated implement contrast, if adjacent points than the center pixel value is small, the pixel position is 0; otherwise it is. The 3 × 3 binary field consisting of eight points depend on the specific rule changes for the line coding, you can get the LBP value point, this value reflects texture information in the region.
Paper completed PCA face detection and face region feature vector dimensionality reduction processing for wireless local area of the face image, which can operate LBP described as follows: p PcaFeature used to describe the dimensionality reduction Haar feature of its neighboring points; Dimensionality reduction Harr feature is used to describe the current pixel and its neighborhood point difference formula as follows: 1, LBP is based on visual characteristics of wireless sensor network identity adaptive recognition target equation can be described as: Where, N is used to describe the number of signatures in the face of the average face; Label used to describe the average face by trained label.

Simulation Analysis
In order to verify the proposed face recognition LBP wireless visual sensor network identity effectiveness of adaptive recognition algorithm needs to implement the relevant experimental resolution. Experiment traditional gait characteristics distinguish the characteristics of identity as a comparison algorithm simulation analysis.

Database Settings
Here in are derived from experimental data using UATV database, experiment database 300 contains images, depicted in Figure 1 is part of a database that contains an image of the sample used in the experiment.

Algorithm Recognition Accuracy Analysis
From a total match score, the average recognition rate of two aspects of the implementation of this algorithm to identify test accuracy.
Respectively, using this algorithm and the traditional algorithm for visual identity of wireless sensor networks in the implementation of adaptive identification, cumulative scores can be used to obtain matching Figure 2 will be described. Analysis Figure 2 shows that the use of cumulative matching score algorithm was 94.2%, the cumulative match scores traditional method is a 76.4% improvement compared to the traditional algorithm cumulative match scores improved 17.8%. This is because This article describes the algorithm effective facial features, improved identification correct classification rate to verify the accuracy of the algorithm.
To further verify the accuracy of the proposed method, the average recognition accuracy of the algorithm and the traditional algorithm were compared and analyzed, the results shown in Figure 3. Analysis can be seen in Figure 3, the algorithm recognition accuracy rate of 95.4%, traditional recognition algorithm 62.7% accuracy rate. Improved algorithm and traditional algorithm, using the average recognition accuracy of this algorithm a 32.7% increase compared to the traditional method, and the average recognition accuracy of the algorithm curve is more stable, indicating that the proposed method provides high recognition accuracy, but also has very high stability.

Algorithm Efficiency Analysis
On different types of child face image database, respectively algorithm and traditional identity recognition algorithms, efficiency of two algorithms were compared, the results shown in: Analysis can be seen in Figure 4, for the same experimental sample, the algorithm time required for 2.4s, the time required for traditional methods of 6.3s. Using the time required for this algorithm compared to conventional algorithms to shorten the time required for the 3.9s, illustrated with traditional algorithms are compared, the higher the efficiency of the algorithm identification, is further evidence of the effectiveness of the algorithm.

Conclusion
In this paper, LBP face recognition [11] wireless visual sensor network identity adaptive recognition algorithm using Adaboost algorithm [12] achieved strong association were cascaded devices, the use of high efficiency square integral image characteristics and the characteristics of the underlying method of extraction, after finally obtained Harr people face to face cascade classifier practice tests. Characteristics of the human face in the field of PCA dimension reduction control vector to obtain a low-dimensional features vector, and down to maintain the number of the person's average face. For wireless local area in the face image, given its operating LBP histogram statistics [13], get face LBP histogram, which feature wireless visual sensor networks and human facial feature database matching, complete identity of recognition. The simulation results show that: 1. The cumulative improvement compared with the traditional algorithm matching algorithm scores improved by 17.8%; 2. The average recognition accuracy of this algorithm a 32.7% increase compared to the traditional method; 3. The time required for the algorithm to shorten the time required for 3.9s compared to traditional methods. 4. It can be seen that the proposed algorithm has a high accuracy and to identify efficiency.

Fund Project
2017 Zhejiang Yuexiu University of Foreign Languages Project (D2017001).
Zhejiang Province Educational Science Plan 2020 (University) Research Project, "Construction and Practice of Experimental Teaching and Management System Based on CDIO Model" (2020SCG357).