Face Recognition Of Gabor Wavelet Research Based On Illuminate Transformation

Face identification has become a research hotspot in the current identification pattern and the machine vision field. The purpose of facial recognition is the computer like humen has a ability to select a right face picture to a right man within many different people’s face pictures.In recent years, the technology of face identification has made a great progress, including many excellent ways of face identification. Many face identification systems have a good function. However, the face recognition technology has many problems remain to be solved,ligthing is one of them.This article offers some insights into ligthing problem of the face recognition how deal with the face image under the charge light condition, how draw the feature of the face and how classify the face. The main completed the following work:(1) How deal with the face image preprocessing under the charge light condition.The existing light pretreatment algorithm are discussed, summarizes the advantages and disadvantages. The primary research is that the arithmetic of light treatment based on the Retinex theory, and we developed a new Retinex algorithm. The different between our Retinex algorithm and traditional Retinex algorithm is that we have used gaussian difference filter instead of traditional low-pass filter to remove lighting component in the image when cutting and modifying in the image, then we used the histogram equalization to enhance processing of image. Simulation experiments show that this algorithm is better than traditional Retinex pretreatment effect. It also remove the halo formation of traditional Retinex, so our algorithm has a higher recognition rate.(2) Studying the ways of face feature extraction based on Gabor wavelet.The Gabor wavelet is insensetive to external environment, such as attitude, lighting,expression, and can extracte robustness face feature. The researches show that Gabor phase feature information contains many effective local features. It is insensitive to the light transform. Besides, original image’s gray value, for the features, it shows all the features of image. Making the image contains both local and globle characteristics, we combined the Gabor phase information with gray value information, and a enhanced Gabor phase characteristics method is put forward.(3) This article has carried on the experimental comparison and analysis with other methods.The design of classifier is the last link of face recognition. The common classifer such as SVM, neural network and so on. They are based on the Euclidean Distance to similarity measure, so they are not used to measure the nonlinearity factors, Such as illumination. so we adopt the Nearest neighbor classification method to classify the facial features. Last, themethod of this article makes a contrast between the method on the face database Yale B and CMP PIE with other pretreament feature extraction is analyzed. Experimental results show that the method in this paper can be a better eliminate the influence of illumination of face image recognition, this article method can effectively improve the recognition rate of face image.