Research On Face Recognition Algorithm Based On Cascaded Regression And LBP

Face recognition technology is one of the important contents in computer vision and pattern recognition field. As face recognition is a complex and high dimensional problem, also influenced by many external factors, it is difficult to achieve relatively high standard in recognition rate and recognition stability. In this paper, a high efficiency and high accuracy face recognition process is proposed. Precise face alignment based on cascaded regression is the first step, the second step is face correction and calculating LBP in a local region around every landmark for face modeling, the last step is classifying faces by SVM. The face recognition process in this paper effectively improves the accuracy and efficiency of face recognition, the test on GT face database shows the recognition rate can reach 99%.The main work in this thesis includes:(1) Study deeply on face alignment based on cascaded regression. Cascaded regression algorithm uses cascaded method to optimize the accuracy of face alignment. We propose 3 kinds of calculation methods in this paper to produce LBF which can improve RF algorithm in regression tree node training. Training and testing are implemented on the LFPW data set, and the result confirms that the algorithm in this paper is accuracy and efficiency on face alignment.(2) Propose a face modeling algorithm based on local LBP. Traditional LBP separates the relationship between features and introduces more noise. The algorithm based on local LBP provides better characteristic data for the training of face recognition because it retains the key information of faces while reducing noise.(3) Construct face recognition classifier based on SVM. Compared to Euclidean distance, SVM can find the key distinguishing feature dimensions from face models which can improve the response speed.