Research On And Improvement To Several Face Recognition Methods Under Different Expressions And Illuminations

In the field of face recognition, how to extract proper features has been an important research direction for a long time. Face recognition might not achieve desired results owing to a variety of factors such as various illumination, expression, pose, covering. To solve the impact of these factors is the major task of the researchers working on face recognition. At present many face recognition algorithms have been proposed for alleviating the influence of these factors. But some of them do not perform ideally, or are only applicable to specific data. Thus, it is necessary for us to improve them or to seek more robust face recognition algorithms. Based on analysis of the status of current face recognition, this paper proposes several methods to solve the impact of various illumination and expression. In this paper, we propose several face recognition algorithms. We also conduct a number of experiments to illustrate the validness of the proposed algorithms.In order to alleviate the impact on face recognition of the illumination, we make some improvements to the local binary pattern (LBP) algorithm. We first partition the image into blocks, regard each block as a new pixel, and use the LBP algorithm to deal with the new image. We refer to this new method blocked LBP algorithm. Based on the blocked LBP algorithm, we put the 0 or 1 into different digits of binary according to there different location relative to center pixel. We refer to this method weighted blocked LBP algorithm. Based on the blocked LBP algorithm, we also take two circles with different size to fix the feature value of center pixel. We refer to this method double circle blocked LBP algorithm. Based on the blocked LBP algorithm, we then fix the feature value of center pixel according to the numbers of 0 and 1 all around instead of binary sequence. We refer to this method binary LBP algorithm.In order to alleviate the impact on face recognition of the expression, we propose a program based on the known expression recognition and expression fitting algorithms. This program is suitable for the case that the faces in comparison database have no expressions while the testing face has. This program combines face recognition and expression fitting. Firstly it recognizes the expression of the face, and then it fits out the face without expression according to the recognition result just get, and last it recognizes the face fitted out. Experiments illustrate that this program has good performance.In order to alleviate the impact on face recognition of the combination of illumination and expression, we propose infrared and visible image fusion face recognition method. Infrared image face recognition and visible image face recognition all have their own advantages and disadvantages. In order to learn from each other, we fuse these two recognition method. Experiments illustrate that the fusion method has higher robustness and better recognition performance.Compared to grayscale face image, color face image contains more information. We propose a color face recognition algorithm. This algorithm use principal component analysis (PCA) method to transform color image to a new image exhibited in two-dimensional irrelevant color space, and then recognizes the two vectors corresponding to the two dimensions respectively, and last fuses the two recognitions on score level. To validate the effect of color space conversion by PCA and the score level fusion recognition performance, we design a series of comparative experiments, and have got good results.