Research Of Face Recognition Based On Gabor Transform Combined With LBP Algorithm

At present, feature extraction algorithm and classification methods for image are the important parts of research in the field of face recognition, in which, face feature extraction algorithm is crucial to decide the face recognition. Gabor transform has not only acceptance field model resemblance to human retina cells to obtain minimum uncertainty for frequency and time, but also stronger robustness for outside conversion like sunlight, posture, emotion. However, traditional human feature extraction algorithm based on Gabor transform folds the human face image and multiple scale and multiple direction kernel function for Gabor to obtain Gabor human feature with sampling and cascade. The obtained Gabor feature dimension of human face is high, the identification process wastes time rather without rotation invariance, and performance of human face figure decreases at plane rotation. LBP is used in operator for texture analysis due to algorithm thought is sample, computation complexity is low and discernment is strong, etc. The above-mentioned analyzed and used widely for several years, LBP operator has fortissimo gray invariance and rotation invariance to overcome the problems of rotation shifting and uneven illumination.Based on the above-mentioned problems of Gabor, this dissertation adopts feature extraction method for combination of Gabor wavelet and LBP to research human face recognition to put forward to LBP algorithm with uniform mode for improving the accuracy of human face feature recognition and flexibility of practical operation. The dissertation provides optimization selection and integration for human face feature extracted by Gabor wavelet and LBP to put forward to improved algorithm: feature extraction methods combined with 2D-Gabor wavelet and uniform LBP. This dissertation algorithm makes progress to improve identification rate and decrease data redundancy with flexibility and effectiveness for human face recognition. The main research results of this dissertation includes several parts:1. First, the research background and meaning of human face recognition technique are introduced. Then the existed human face recognition algorithm at home and abroad is introduced to come up with the deficiency of human face recognition technique and problems required to be solved. Finally, put forward the new algorithm, and introduces the main content and structure arrangement.2. This dissertation introduces the common human face recognition system, including human face database, pretreatment and classifier design, retrospect classical human face identification technique, experimental simulation for human face recognition algorithm based on PCA human face recognition. Then, this dissertation makes a detailed introduction for Gabor conversion, including functional introduction, parameter choosing of filter bank, human face feature extraction algorithm based on 2D-Gabor. Finally, this dissertation makes a detailed introduction for LBP. Including LBP operator introduction, uniform LBP algorithm.3. Raises a improved algorithm. That is to say combined feature extraction algorithm based on 2D-Gabor wavelet and uniform LBP. This dissertation researches the influence for identification performance of 2D-Gabor wavelet core window size, LBP partitioning mode and feature dimension. Eventually, This dissertation makes theory analysis and experimental simulation for improved algorithm, to verify the effectiveness and feasibility for algorithm comparing to traditional algorithm.