Yanqiu Li, Fuji Ren, and Min Hu
Expression recognition, DT-CWT, Inverse cloud model, Adaptive weighting matrix, Information entropy
In order to describe the facial expression images in multi-scale better and weaken the influence of the edge information to recognition result, this paper proposes a method for describing the feature of the facial expression which uses the dual tree complex wavelet transform (DT-CWT) and Uniform Local Gradient Pattern (ULGP). To distinguish the contribution which feature information of different directions play in the recognition, and describe the randomness of the testing image, the cloud model is introduced in this paper. It uses the information entropy of features in different directions as the cloud droplet to obtain the membership degree of the testing image and get the adaptive weighting matrix of it. Moreover, at the recognition stage, combined with BP neural network, we obtain the posterior values which the testing image in different directions belong to different categories. Finally, we utilize the weighted fusion method of decision level to realize the recognition of the testing image. The experimental results on JAFFE database and CK database verify the validity of the feature extracting method in this paper, and the feasibility of using the cloud model to get adaptive weights. The system obtains higher recognition rate ultimately.