Based On Support Vector Machine For Color Image Positioning Of The Human Eye And Face Detection

As a special case of generic object detection, face detection can be applicable in the field of fully automatic face recognition system, content-based image retrieval, digital video processing, surveillance systems, human-machine interface and so forth. It's of great importance both academically and commercially.This thesis focuses on the research of human eye pair localization and face detection based on previous localization results in color images. As is a critical step towards face detection and recognition, the positions of eye pair are commonly used for the geometry normalization of a human face image by estimating its scale and orientation. The results of eye localization premise the verification of face region in this thesis. Support vector machines (SVM) are used as the verification algorithm for its excellent performance of classification. In particular, main contributions of this thesis detail as follows.Firstly, based on the prior knowledge that two eyes are in the face, i.e. eyes are surrounded by skin, skin segmentation can be applied to reduce the search region. Here, a method of segmenting the skin likelihood grey scale image is proposed based on the hysteresis thresholding technique. Compared with those using a single thresholding value, it's more effective.Secondly, an algorithm of eye localization is proposed. To reduce the amount of SVM verification, the algorithm utilizes chrominance, luminance and gradient information of eyes to further determine candidate regions. For those error-detected regions, geometry information of eyes, such as the angle of eye pair and the ratio between the distance of two eyes and the width of the outer rectangle of detected region, can be the rules of removal. Then, more precise locations of eye centre can be found by using projection functions.Thirdly, some eigenvectors are extracted for training and predicting face images by casting singular value decomposition (SVD) on face data. Compared with an eye, a face image is larger and more variable. Thus, it's time-consuming. To solve this problem, scalar normalization and SVD are used for extraction of features.Experimental results show that the method of this thesis is quite useful and practical.