Research On Algorithm Of Facial Expression Feature Extraction And Recognition

Facial expression recognition (FER) is an important part of biological recognition technology. It is widely used in the fields of intelligent human-computer interaction, robotics, medical care, entertainment, virtual reality technology, etc. Facial expression is always passing a lot of information when people communicate each other, so it plays a very important role in daily communication. In recent years, facial expression recognition technology has been paid much more attention and has become a research focus in the information processing flied.In this paper, the key technologies of feature extraction and recognition of facial expressions are researched. A fuzzy fusion approach for facial expression recognition based on Gabor transform optimal channels is proposed. Three main contents are as follows in this paper:①Expression images are preprocessed firstly. It includes human eyes location, key region cropping, rotation, scale normalization, illumination compensation, etc. A method that combines geometric characteristics with Hough transformation is used to locate human eyes and precise location of the eyes is obtained.②Multi-scale and multi-directional properties of Gabor wavelet kernel function are analyzed in detail. 30 Gabor filters are designed by choosing its five scales and six directions to extract texture features of face expression images. Because of high dimension features of Gabor wavelet transform, a reducing dimension method that combines non-uniform sampling with two dimensions Principal Component Analysis (2D-2DPCA) is proposed and required expression feature vectors are obtained.③Because different channels have different contributions to the recognition rate, a fuzzy integral fusion approach for facial expression recognition based on Gabor transformation optimal channels is proposed. Firstly, four optimal channels are chosen according to proposed four optimal channels selection principles. Next, the features that extracted based on each optimal channel are input into BP neural network to recognize facial expression. Finally, the expression recognition results are fused based on multi-classifier combination with fuzzy integral to get final recognition result. The experimental results on JAFFE database illustrate that the recognition rate of the proposed algorithm is 97.65%. The contrast experiments are used to verify the effectiveness and superiority of the proposed algorithm.