The Research On Facial Expression Recognition Based On Multi-scale Local Gabor Binary Patterns

Facial expression recognition is a very challenging and meaningful issue, itrelates to computer vision, psychology, physiology and some technology in patternrecognition and image processing. With the continuous improvement of computerperformance and low-cost digital image capture technology, facial expression visionhas been experienced a rapid development, in which it is necessary to recognizeexpression in a single static image for many applications. However, as a single staticimage contains less expression information than image sequences, facial expressionrecognition from a single static image is more difficult. In order to achieve betterresults, this paper brings out a new method based on Multi-Scale Local Gabor BinaryPatterns algorithm. The main works of this paper are as follow:(1) In face detection, this paper introduces face detection based on the Haar-Likecharacteristics and theAdboost learning method, and we extract the facial region.(2) In image preprocessing, this paper takes some operations to preprocess theface image, like size normalization, removal the noises and grayscale equalization onthe facial region. Then mark out the main facial expression region, laid a gooffoundation for feature extraction and classification.(3) In feature extraction, this paper introduces feature extraction based onMulti-scale Local Binary Patterns, and then introduces Multi-scale Gabor analysis toachieve globally expression information. We combine those methods and bring outvethe Multi-scale Local Gabor Binary Patterns (MS-LGBP) features, which is bothlocally and globally informative.(4) In classification, based on BP neural network, this paper realizes a classifierand carries out experiences on JAFFE expression database. The result shows withMulti-scale Local Gabor Binary Patterns (MS-LGBP) features, the expressionrecognition achieves good results.