Research Of Face Recognition Based On Gabor Transform Algorithm

As a challenging problem in the field of multi-disciplinary, facial Recognition Technology is widely applied in many aspects. In-depth study and solve problems of face recognition having great academic research and practical application value, which is beneficial to promote the development of related disciplines. Advantages and disadvantages of face recognition algorithm are most likely to depend on the human face feature extraction and its precision. Good face features can use simplest classifier to achieve best effect.Gabor Transform (GT) is a very good feature extraction algorithm, having a striking similarity to the accept model of human retinal nerve cells. Because of the traditional feature extraction method based on GT has many deficiencies. Therefore, this paper adopts a deformation of the GT, named the Circular Symmetric Gabor Transform (CSGT), used in the study of human face feature extraction. CSGT not only has GT general nature, but also has advantages of small redundancy and Rotation Invariance. In order to make full use of the characteristics of CSGT and further improve the accuracy of the facial feature extraction and flexibility actual use, this paper optimize the CSGT extraction of facial feature selection and fusion, and propose two improved algorithms based on CSGT:CSGT Texture Statistical Feature (CSGT-TSF) algorithm and CSGT Weighted Multichannel Texture (CSGT-WMT) algorithm. Finally, paper improves the dimension reduction algorithm and effectively realizes the face recognition. Main research results of this paper include the following parts: 1. The detailed introduces and implements of the Classic Face Recognition System. This paper first uses Principal Component Analysis (PCA) to realize face recognition. Then, the paper detailed introduces the GT, including:kernel function, parameter selection, feature extraction. Finally, the paper uses the improved algorithm based on GT+PCA to extract facial features, and simulate and implement the GT face recognition.2. The CSGT abilities are verified by theoretical and experiments. As the previous research based on CSGT is not enough, the system theoretical research of this paper is based on CSGT. Experimental simulation implements the facial feature extraction based on CSGT.3. Proposed CSGT Texture Statistical Feature (CSGT-TSF) algorithm. The data quantity of the CSGT transform feature is five times then the original sample. It is necessary to carry out CSGT characteristics optimization. Therefore, this paper combines Texture Statistical Feature (TSF) and CSGT, then proposes CSGT-TSF algorithm. Using statistical feature extraction method, this algorithm effectively solves the identification of "False Registration Problem". Combined with sub image processing way, this algorithm fully gets CSGT robust local texture feature. Finally, this algorithm extracts and constructs important identify face image multichannel texture feature. Experimental results verify that CSGT-TSF can effectively reduce the feature dimensions of CSGT, identify performance is better than CSGT, GT and other classic algorithms.4. Proposed CSGT Weighted Multichannel Texture (CSGT-WMT) algorithm. Based on CSGT-TSF algorithm in the fusion stage, in order to overcome the defects of the Cascade algorithm, distinguish the importance of the texture statistical characteristics in different channels, CSGT-WMT facial feature extraction algorithm is proposed in this paper. Using Adaptive Weighted fusion the algorithm dynamically adjusts the weights of characteristics in different channels, gets the most discrimination feature. Compared with CSGT-WMT, CSGT-TSF, CSGT and other traditional algorithms, experimental results show CSGT-WMT having high recognition rate, short recognition time, good robustness in illumination, posture and local deformation interferences, flexible to adapt to different face databases, good classification performance and practical application ability.5. Proposed face recognition algorithm based on Weighted PCA (WPCA) dimension reduction. This paper improves the classical PCA dimensionality reduction algorithm to WPCA, thus puts forward CSGT-TSF+WPCA and CSGT-WMT+WPCA, two joint face recognition algorithms. Experimental results show that joint algorithms have good dimension reduction effect, strong anti-interference ability and high recognition rate.