Face Detection Based On Convolutional Neural Network And Improved Support Vector Machine

Face detection is the premise of face related research, and is the basis of pattern recognition and computer vision. However, one key challenge of face detection is the large appearance variations due to some real-world factors, such as occlusion, viewpoint, extreme illuminations and expression changes, which lead to the large intra-class variations and making the detection algorithm not robust enough. Therefore robust face detection algorithm has practical application value but equally exits a big challenge.Face detection is mainly divided into two steps:feature extraction and feature classification. The research work in this dissertation is also carried out in such two aspects. The main work of this dissertation is as follows:(1) Feature extraction:Most approaches are generally based on Haar, PC A, LBP and other hand-crafted facial features. In this dissertation, we present a face detection algorithm based on features learnt using convolutional neural network (CNN) automatically so as to explicitly capture various latent facial features. Firstly, in order to improve the detection speed of the system, we train an Adaboost background filter which can remove the background most quickly. Then we use the CNN to extract more distinctive features for those face and non-face patterns that have not filtered by Adaboost. Finally, support vector machine (SVM) is used to classify the extracted features as face or non-face. Experimental results show that this algorithm combines the advantages of CNN learning characteristics and SVM classification, improving the algorithm performance.(2) Feature classification:Partial occlusion has brought a series of problems for face detection, such as lead to large intra-class variations between face images and making the classification difficult. In order to solve the problem of partial occlusion, we propose a locality-sensitive support vector machine using kernel combination (LS-KC-SVM) algorithm. First, we employ the locality-sensitive SVM (LSSVM) to construct a local model on each local region, which can handle the classification task easier due to smaller within-class variation. Then, in order to solve the partial occlusion, we combine the global kernel and local kernel, thus it can measure both detailed and rough similarity. Finally, we apply this combination kernel to LSSVM. Experiments prove the good results of it.(3) Algorithm combination:The proposed feature extraction and feature classification algorithm are combined to construct a complete robust face detection system. For the combination of the two algorithms, unlike the traditional way to use a network to extract the whole face feature, but to fit the characteristics of the LS-KC-SVM algorithm.