Facial Expression Recognition Method Based On LBP Multi-features Fusion

As face recognition gradually becomes more popular and widely-used, more and more people begin to focus on computer recognition on facial expressions.Face detection and feature extraction are of great significance in the area of pattern recognition.Currently, as for face detection and feature extraction in facial expression recognition, many classical algorithms have emerged,including linear and nonlinear feature extraction algorithms. Among them, local binary pattern has received a lot of attention and research.Core contents and directions of this paper can be summarized in the following three aspects:Based on the existing Ada Boost, traditional Ada Boost algorithm merely focused on the issue of minimum error rate. The face detection method based on the combination of LBP and the Cost-sensitive Adaboost is proposed. This paper adds cost-sensitive learning model to Ada Boost algorithm. Firstly, LBP feature extraction is conducted on facial images. Then, in the phase of training classifier,cost-sensitive learning is introduced. As for samples from different categories,sample weights are updated based on the different costs caused by error dividing.Improved algorithm focuses on minimum cost of classification instead of minimum error rates.As for CS-LBP algorithm, horizontal symmetrical components may pose negative effects on image recognition. The new facial expression recognition algorithm based on the combination of Enhanced Center-symmetric Local Binary Pattern and Embedded Hidden Markov model is proposed. First, enhance the priority levels along the image diagonal directions. Then reduce the negative effects of horizontal component on images. Lastly, conduct facial expression recognition combined with EHMM. Through experimenting on CK and JAFFE face database, results show that recognition rate of this algorithm is obviouslyhigher than that of the other local binary pattern algorithms.In view of computational complexity and high dimension problems of five dimensions and eight directions of the Gabor transform. A feature extraction algorithm based on the Gabor feature of enhanced center-symmetric local binary pattern is proposed. We conduct improvement on Gabor, and feature extraction combined with ECS-LBP. Firstly, use Gabor to extract the amplitude image expression features in five dimensions and eight directions. Integrate the amplitude images of the same direction in different dimensions. Then use enhanced CS-LBP to further extract features. While reducing dimensions, more feature information of expression images can be retained.