自适应加权完全局部二值模式的表情识别

Facial expression recognition based on AWCLBP

为了有效地提取局部特征和全局特征以提高表情识别的性能,提出自适应加权完全局部二值模式(AWCLBP)的人脸表情识别算法。首先对人脸表情图像进行预处理分离出表情子区域,与此同时生成表情子区域的贡献度图谱(CM);然后对表情子区域和整幅表情图像进行完全局部二值模式变换提取3种特征(差值符号特征CLBP_S、差值幅值特征CLBP_M、中心像素特征CLBP_C)并连接3种特征生成级联直方图,并根据CM对表情子区域的级联直方图进行加权与整张图像的直方图进行融合;最后用卡方距离和最近邻方法进行分类识别。该算法在JAFFE库上进行了实验并和LBP、Gabor小波、活动外观模型进行了比较,验证了该算法的有效性。

In order to improve the performance of facial expression recognition by extracting the effective local features and global features, an algorithm for facial expression recognition based on adaptively weighted compound local binary pattern （AWCLBP） is proposed. First, the facial expression sub-regions are isolated by a preprocessing step. Then, the contribution maps （CM） of facial expression sub-regions are computed; Second, a compound local binary pattern （CLBP） extracts expression sub-regions and the global facial expression image and then cascade histograms are generated by connecting the histograms of the three features of the image and the one expression sub-regions that is weighted accord- ing to the CMs. Finally, the weighted cascade histograms are classified and recognized by using the chi-square distance and the nearest neighbor method. Experiment results on the facial expression database of JAFFE show that the proposed algorithm can be applied to achieve a higher recognition rate than other algorithms, such as, LBP, Gabor wavelet and the active appearance model.