Research For Mixed Noise Face Recognition

Due to its stability, security and non-invasive, face recognition has always been the hot spot of the scholars study. Compared with other human biological characteristics identification, face recognition has a friendly way, convenient sampling, no contact with many advantages, such as and, face feature has a strong own stability and individual differences, is the ideal basis for authentication, so the research on face recognition research has important academic value and broad application prospects. But so far most studies are only for the pretreatment and feature extraction for single noise face images, etc. The actual image signal is affected by Salt and Pepper noise, Gaussian noise and environmental noise, etc.The pretreatment is a key part of the face recognition, consist of three aspects, including noise reduction and normalization. And noise reduction methods mainly include soft threshold,hard threshold and semi-soft threshold method. However, these methods still exist two main problems: one is threshold function continuity; the other is the constant deviation between the estimating wavelet coefficients and the wavelet coefficients of noisy signals(as “constant deviation problem”). Aimed at deficiencies of these methods, Scholars have proposed many improvement programs, which haven't completely solved those problems, but one of constant deviation and continuity function. In order to overcome the above problems, it focuses on the pretreatment and feature extraction of mixed noise(Salt and Pepper noise, impulse noise and Gaussian noise) face images. An improved threshold function is proposed to completely overcome two main problems of traditional methods.Eye location is one of the most important process of all the face recognition system,which will effect the final feature extraction and recognition rate. Eye location is the basis of the other feature points including the nose and mouth, and the face feature points are assumed to be given in most recognition experiments, or in conditions that it allow users to a certain degree of interaction, and it is lack of a common and perfect eye location template. It combined the projection methods to quickly and accurately position the eye in the case of face poses, illumination changes and wearing glasses; Face image feature extraction for face recognition robustness and efficiency has a decisive role, a friend in the process of face recognition, feature extraction is one of the most important steps, and the stand or fall of feature extraction and feature extraction method and the early stage of the image preprocessing work is closely linked, in order to extract the effective classificationcharacteristics, this paper combined 2DPCA-LDA and ILPP to get more accurate extractions of effective facial features according to the needs of different sectors, so as to improve recognition rate and have higher practical value; Finally, the processed face images will have been identified and classified by SVM methods.