| Summary: | 碩士 === 中華大學 === 資訊工程學系(所) === 96 === Identifying human facial expressions on still image is a challenging task in computer vision research. We have presented a multi-feature model capable of recognizing facial expressions. The proposed model adopts wavelet transform based two-dimensional principal component analysis (W-2DPCA), which performs better than traditional PCA and local binary pattern (LBP), which is efficient for analyzing the texture of a image. In this paper, we further proposed
an edge-based LBP (ELBP) and improved the recognition accuracy. A weighted combination of 2DPCA and ELBP features is input to the decision directed acyclic graph (DDAG) based support vector machine (SVM) classifier. Thus, the mixed-feature model has the advantages of both features. The proposed method is measured on six facial expression databases, AR, CohnKanade, JAFFE, GUR, Nimstim and Yale, and is compared with other methods. Experimental results indicate that the proposed model is feasible and the identification rate outperforms other methods.
|
|---|
