Facial expression recognition research based on blocked local feature

Focus on the key feature extraction and selection of facial expression recognition, this paper firstly extracts the global Gabor features and facial key expression as local features. Then centering, rotating and scaling every feature block with Procrustes analysis to reduce the effect of position and size inconsistent. At last, reduce the dimension of global Gabor features and local features with PCA algorithm and combine them to fused features. Experimental results show that whether the single features or fused features, blockProcrustes can obviously improve the expression recognition performance in most cases, especially increase the stability and maximum of recognizing accuracy.