Facial Expression Analysis Based On Local Binary Patterns (LBP)

Feature extraction methods based on LBP (Local Binary Patterns) features have been applied to facial expression recognition, and demonstrated LBP features'superiority for facial expression recognition. Aimed at the subject of facial expression analysis based on LBP features, this thesis carried out in-depth study, involving facial feature points detection, LBP feature extraction, and expression classification, etc. At last a new feature extraction method was proposed based on Local Binary Patterns.From the view of facial image analysis, there is much redundant information in the extraction from the whole face. How to select valid LBP feature in a suitable way is the important issue of the application of LBP operator. It is a major breakthrough to improving LBP method using the main information of local patterns more effectively and rationally. To achieve the solution of this problem, the extraction methods of local LBP feature is mainly researched in this thesis, which arranged as follows:1. Classical LBP feature extraction methods are in-depth researched and studied. It is a common method of extracting global facial image's LBP feature. Filter the whole facial image's pixels with LBP operator, divide the image into blocks, then extract each block's LBP feature and construct histogram feature vectors, which include redundant information. This thesis in-depth researches and studies on classic LBP feature extraction method, and summarize the disadvantages of it.2. Feature extraction method based on LBP features of local area is proposed. For facial expression, the eyes, eyebrows, nose and mouth are very important, and the related regions contain more useful information. Therefore, it should be able to enhance expression features to give these regions relatively large weight. Moreover, in the phase of features extraction, extract LBP features of the facial points'region and construct feature vector, and then the most distinguished features can be reasonable extracted. We constructed experiments based on the combination of the face division method of LLBP and Gabor wavelet feature extraction method.3. Feature extraction method based on difference of LBP features and difference's LBP features are proposed. For the same individual's expression image, its organs'distribution, size and shape, such as eyes, nose and mouth, have difference with that of neutral image. These differences are facial expression's nature information. The concept of difference of LBP feature and difference's LBP feature and extraction method are introduced in detail. Then, experiments are conducted respectively, followed by analysis and comparison. Experimental results show that differences of LBP features are more effective in describing the differences between expression image and neutral image.4. Feature extraction method based on difference's LBP features of local area is proposed. This thesis combined difference's LBP features with the method of LLBP, Filter the difference image of expression image and neutral image with LBP operator gain the difference images LBP map, then use LLBP method, divide it into several expression sub-region, extracting each block's LBP histogram of each expression sub-region to construct feature vector.