Research Of Facial Expression Feature Extraction Method Based On LBP Histogram

Facial expression is an important research topic in emotional computing, intelligentcontrol, computer vision, image processing, and pattern recognition. The static facialexpression recognition is that recognizing facial expression in the static image. Commonly, thebasic human expression is defined as seven categories, the researchers used PCA,LBP andGabor transform to extract and classify features, but the result is insufficient and inaccurate, itaffect the recognition rate of the facial expression. So we need to research the better method toextract and classify features.The main content of this thesis is the method of facial expression feature extraction. Inorder to overcome the disadvantage of traditional Gabor filters, whose high-dimensionalGabor features are redundant and global features representation is insufficient. In this thesis,the Gabor multi-orientation features are fused and combined with block histogram to extractfacial features. In the method, all important parts are distinguished in different expressions,proposed a facial expression description model based on prior knowledge and defined thecertain detective directions in selected regions based on Gabor filter direction features. TheGabor filters are implemented to extract features from these regions in multi-scale andmulti-direction, compared the Gabor magnitude with gray to retain the stronger one, convertedfrom8orientation Gabor feature to the8binary code and then the features are fused in thesame scale. Finally, based on the facial expression description model, used LBP histograms toextract feature from fused features. After fused the images, we can reduce the featuredimension from184320to23040. We calculated the weight of the selected regions by usingthe result of feature detection. Our method can improve the accuracy of the feature extraction,highlight the important features effectively, and reduce the feature dimension greatly.Testing and analysis of method and system in JAFFE facial expression database, theproposed method can effectively extract facial expression features. The system can effectivelycomplete facial image pre-processing, facial expression feature extraction, facial expressionfeature weight calculation and facial expression recognition. The system proves to have a highrecognition rate in JAFFE.