High-dimensional Feature Based Face Verifacation Using Support Vector Machine And Adaboost

Face recognition is fundamental to a number of significant applications that include but not limited to attendance, video surveillance and content based image retrieval. The general problem of face recognition is actually comprised of two related problems: verification and identification, and the application of the former is more extensive. But some of the challenges which make this task difficult are variations in faces due to changes in pose, illumination and deformation. This dissertation proposes to construct a high-dimensional(e.g.100k-dim) feature which contains more information of face to overcome these difficulties. And the face verification is realized in the mine attendance system using support vector machine and Adaboost.The common feature of face always just contain a part of the information of face. And the accuracy of face verification with common feature is not acceptable, because one feature has its own characteristic. The high-dimensional feature can improve these problems but at the same time high-dimensional feature will bring more time, calculation and storage. To deal with these problem, this dissertation do these work as follow.(1) To construct the high-dimensional feature containing more information of face. Firstly, dense facial landmarks is found with a recent face alignment method and the image is rectified similarity transformation based on five landmarks. Then multi-scale patches centered around each landmark is extracted. Finally, the high-dimensional feature is concatenate from the all patches.(2) To realize the face verification using the SVM and the Adaboost. The SVM uses the feature whose dimension is reduced by PCA to realize face verification. But the calculation of PCA is huge, so the verification method using the Adaboost is adopted. The Adaboost can learn effective features from a large feature set and boost the weak classifiers which is constructed based on one of the selected features into a stronger classifier. Both methods are tested on the FERET image sets, and the accuracys of the two methods exceed 97%.(3) Face verification system has be realized in the mine attendance system, and the coal miner face data(CMFD) which contains 2753 miners is completed. The CMFD including the faces in different posture, expression and coal ash has great value in research and application. This dissertation focuses on faces with coal ash and slag. It reached 86.95 percent success rate by one-time authentication using the above method.This paper studies the characteristics of high-dimensional feature based on multi-scale transform and facial landmarks. The high-dimensional feature brings the positive impact to face verification. And the face verification methods using the SVM and Adaboost algorithm based on high-dimensional feature are tested on FERET face database. Finally, the face with coal ash and slag is studied, and the face verification methods in this paper has been applied to coal mine attendance system.