Color Image Based On Face Recognition

Face recognition is a research hotspot of pattern recognition.Generally.the traditional face recognition method, extracting single feature in gray or brightness images and then calculating it. ignores color information. Recent studies have showed that color plays a significant role in face recognition, extracting feature by adopting the color image and selecting the appropriate color space brings better recognition rate than using feature from gray or bright image only. This thesis studies preprocessing methods of color space and image, feature extraction and classification methods, respectively.The main achievements are as follows:(1) Discriminant color space model is studiedFirst, the color space applied to the color face recognition is analyzed. Then two kinds of Color space normalization methods are studied:the within-color-component normalization technique (CSN-Ⅰ) and the across-color-component normalization technique (CSN-Ⅱ). Finally, the discriminant color space method is studied. This method is to construct discriminant color space model by training and then transform the image from RGB color space to the discriminant color space. In this thesis, the AR databases and self-built color databases NUST_RWFR have been applied in the experiment. The experimental results show that, compared with other color space, the discriminant color space has achieved a higher recognition rate and improved the color facial performance.(2) The feature extraction methods using Gabor wavelet and Local Binary Patterns (LBP) are studiedFirst of all, two face recognition methods, the Gabor wavelet and LBP operator are studied. Then both methods are applied in the color space, and the feature extraction method for color images based on combination of Gabor wavelets and LBP operator is proposed. Our algorithm not only prevents missing important classification information due to the use of uniform down-sampling for dimension reduction of the Gabor amplitude features, but also improves recognition rate effectively by using histogram for dimension reduction. Finally, the experiments on NUST_RWFR databases and AR databases are carried out respectively. The method based on combination of Gabor wavelets and LBP operator turns out to be more effective than traditional Gabor filter and LBP algorithm, and face recognition using color image performs better than using the gray images. (3) Robust sparse coding identification method combined with multiply features is proposedFirst, the linear regression and sparse representation theories and methods are studied. Then both methods are applied in face recognition. The robustness for face recognition is further improved by using the maximum likelihood maximum theory. Robust sparse coding method combined with the feature of LBP and Gabor+LBP feature is put forward and is applied in face recognition. Experiments are carried out in NUST_RWFR databases and on AR databases, respectively, and the results show that this method obtains higher recognition rates compared with traditional classification methods.