Research And Implementation Of Face Recognition Based On Gabor Features And Neural Network

Recently, along with the rapid development of our economy and society, the security has become one of hotspot problems. As one kind of biometric technology, face recognition technology has stability, convienience, and uniqueness features, and which is concerned by the majority of scholars. How to effictively improve the recognition rate and ensure the information security is particularly important in our society. The face detection and location, feature extraction and matching authentication are the three key technical problems in the face recognition. Based on the color face images under the natural state with the background, that collected by the camera equipment, the face recognition is studied. The main work of the paper includes:(1) Aiming at the distinctive differences between human face skin and background color, a skin segmentation algorithm is proposed based on YCb Cr color space. The threshold of the Cb and Cr two color images are extracted by utilizing an improved Fuzzy Entropy Theory, and then the skin and non-skin regions are separated by using fuzzy IF-THEN rules. According to the shape of facial features, the detction and location of face are realized. The simulation results show that the accuracy of the face detection has been greatly enhanced.(2) Aiming at this drawback that the tranditional feature extraction method has higher dimension and is easily affected by illumination, expression and other factors, an improved Gabor feature extraction algorithm is presented. Due to the special color information of the eyes and lips in the chrominance components, the exact locations of feature points can be obtained and extracted. And then, the position of the nose is roughly determined. Through combining with Sobel edge detection, the feature points of nose are extracted. On this basis, a set of Gabor filters can be designed to automatically obtain the local texture characteristic nearby the feature points. The simulation results show that the extracted Gabor features not only have the lower dimension, but also has good robustness.(3) Based on the processing capacity, that the BP network has, of complex problems, a novel neural network classifier set is constructed to classify the features. The designed classifier set is composed of several single classifier, and its output and input are in accord with the categories of face feature. By using this set, the different haracteristics types can be trained and tested. The simulation results show that the novel constructed neural network classfier set has higher recognition rate.(4) Based upon the hybrid programming of C#, MATLAB languages, and Visual Studio 2012 development platform, a face recognition system is developed. The system can reveal the whole process of face recogniton, and realize the certification of the face in real time. The results of the system show that the proposed approaches can also acquire the features of the face in the variational illumination, facial expression, angle conditions. The similarity of the same person in different periods also reached more than 90%. The designed system has faster running speed and higher accuracy.