Research And Realization On Face Recognition Technology

The human face recognition is a subject with great theoretics and application value. With the development of society and the improvement of science and technology, it is urgently needed for convenient, reliable and automatic status discrimination. Thus face recognition technology has become the focus of machine intelligence research field. The significance of face recognition is not only to promote the development of image operation, pattern recognition, cognition science, psychology, physiology and other interrelated subjects, but also to meet some practical needs such as identity confirmation and the search based on contents.Aiming at some unsolved problems on face recognition field and connectting with other people's researches, we do some researches on the three aspects of composing a face recognition system: face detection, feature extraction and face recognition. This paper mainly includes the following nubs:1, Modified and improved the tranditional face detection method based on skin model. Firstly this mothod detects dubitable face areas by skin model, and then this mothod uses eye-template to search these areas for eyes' accurate position. Experiments show that this method has high practical value.2, Proposed a feature extraction method based on local feature analysis. Firstly some pre-processing is done on the face image, secondly this method extrats the local geometric features of the apparatus on face according to the apparatus' configuration, color and position relation. Eyes' position is the benchmark of these features. Experiments show that this method can extract facial features effectively.3, Introduced a rule of order and mark in the stage of feature matching. By this rule, every feature is endowed with a particular weight value, and all the training swatches are marked and ranked from high score to low score. Besides, this paper proposed an approach of hiberarchy identification. This approach separates the features into several levels according as features' importance. Experiments show that this approach is useful in feature matching. 4, Proposed a face recognition method based on whole principal component analysis (PCA) and local principal component analysis. In this method, PCA is done twice, one is on the whole face and the other is on the eye areas. The two results with each weight value are combined to form the final result. Experiments show that thiscombined method exceeds the tranditional PCA.5, Proposed a kernel principal component analysis (KPCA) method for facerecognition based on wavelet decomposition and smoothing filtering. In this method, the low-frequency component of face images can be retained and the high-frequency component of face images can be weakened by wavelet decomposition and smoothing filtering. The non-linear feature of face images can be obtained by kernel principal component analysis. Experiments show that this method has higher recognition rate than usual KPCA methods.