Design And Implementation Of Face Recognition Based On Hidden Markov Model

Face detection and recognition technology, as one of those computer visual research areas which attracts increasing attention in recent years, has been widely applied into the field of security. By using a face, an intrinsically unique biology feature, the process of detection and recognition achieves a safe, reliable and effective identity confirmation.This thesis studies technologies involved in face recognition, such as: image processing, facial detection and recognition. The main research work in this thesis can be summarized as follow:it analyzes image processing methods used in this thesis, including grayscale, image resizing and filtering, with a focus on detection and elimination of impulse noise. By analyzing examples, this thesis finds a flow in traditional noise detection method, which could lead to an error in acceptation of impulse noise and increase in processing time. An improved approach has been proposed by introducing an average distance into the traditional method. Because detection operation is carried out frequently with a strict requirement on real-time performance, this thesis comes up with a solution which combines haar feature based cascade and hidden markov model(HMM) in the realization of face detection and recognition. On the research of face detection, haar feature based cascade is used to extract human faces in a given image, which offsets the slow speed of HMM in classification by taking advantage of real time feature of haar feature based cascade.In the aspect of face recognition, this thesis focuses on discrete cosine transformation (DCT) feature extraction method used to provide HMM with input data and finds out that when directly executed, DCT generates image features with a huge amount of data, increasing time needed to perform recognition. After an analysis of wavelet transformation, wavelet based data compression method has been introduced into the process of facial feature extraction to obtain approximate images. And then, DCT features are calculated, ensuring a nice real-time performance of recognition system.Finally, using the approaches designed in this thesis, a face recognition system is realized based on VC++ integrate development environment. Experimental results achieved on a collected face database prove the feasibility of proposed methods.