Human face recognition based on a single front view

Abstract

Human face recognition is one of the most useful techniques for identifying or authenticating a person. Although research on this topic has been conducted for more than twenty years, many problems still remain, and better techniques for facial feature detection and face recognition are needed. Therefore, the objectives of this thesis are to devise and develop efficient methods for preprocessing facial images and recognizing human faces. In this thesis, different approaches for human face detection, facial feature extraction and human face recognition are reviewed. Human face detection and facial feature extraction are the preprocessing steps for automatic human face recognition. Their accuracy will directly affect the performance of the recognition system. However, since the location of a face, its facial expression and lighting conditions in an image are unknown, and considering that its size and orientation may be different, the recognition procedure is difficult and computationally intensive. Thus, human face recognition is a challenging research topic. In this research, we propose a fast approach based on valley field detection and a modified fractal dimension to extract an eye pair in a complex background, which can then be used to represent a face region. Instead of searching the whole image space to determine the scale of a face, only possible eye pairs as detected by the valley field and their local properties are investigated. These possible eye pairs are then identified by means of the modified fractal dimension. Furthermore, in order to improve detection reliability, uneven lighting conditions on the two halves of a face are normalized by means of a histogram technique. The corresponding average fractal dimensions of the binariied eye-pair regions and the face regions are then used to verify whether the eye pairs selected are valid. Human face recognition techniques focusing on whole face and facial features such as the eyes and mouth have been proposed. Due to the fact that different facial regions have different degrees of importance for face recognition, a new modified Hausdorff distance is proposed. This distance measure incorporates the a priori structure of a human face to emphasize the importance of facial regions. The face recognition technique proposed in this thesis is computationally simple and can provide a reasonable performance level.