Human Face Recognition System Based On Extended Lbp Feature Research

Face recognition has been a hot research problem of pattern recognition and machine vision. The face is a kind of biological characteristics, directly, uniqueness, convenient acquisition, but because of the fluidity of the face and vulnerable to external factors, makes the subject of face recognition is very challenging. Existing algorithms is significantly poor robustness to environmental changes, especially compared with the face recognition capabilities, automatic face recognition performance on the environment and attitude change adaptation is low, the majority of the existing methods case is to be detected face collection in the specified environment, which obviously does not meet the practical requirements. This paper studies and improve the classic face recognition algorithm, including image preprocessing, characteristic collection, feature reduces, etc, and the initial realization of a video-based face recognition system prototype. The main work is as follows:1. Research LBP operator of several extended form, including the unified model, the rotation invariant unified model and improve the three-valued form of LTP model. Through research, we can see the LBP operator has two features:rotation invariant and grayscale robust. It is very important to face recognition, LBP also has the ability to describe the local and global features of the image, This is also very beneficial to face recognition; the LBP value of this article will be an extended form of LTP mode is applied to face recognition, compared to the traditional LBP model for the local illumination variations and noise, a higher level of robustness.2. In-depth study of the dimensionality reduction algorithm of the original model of LBP. Some of the defects that exist because of the the LBP unified model itself, so we use the LBP sub-model generated by the dimensionality reduction instead of the LBP unified model; This paper compares three dimensionality reduction algorithm (locality preserving projection (LPP), principal component analysis (PCA) and linear discriminant analysis (LD A)) in the field of face, and ultimately selected the LPP mode in building the LBP sub-mode.3. Research face image preprocessing methods, especially the face correction. This paper show a method:first through the average of synthetic exact filter (ASEF) to obtain the location of the human eye then change the face rotation by affine transformation, and finally to the frontal face of the approximate standard. ASEF given the precision of the human eye is much higher than the Haar-like+Adaboost method, and it’s more robustness for wearing glasses or the block face, the most important is its very fast, almost to the detection face at the same time be able to locate the human eye.4. Open source library based on OpenCV Constructed a prototype of a video-based face recognition system. In the above algorithm, the actual test.