Abstract: Put forward a combined with improved LBP（Local Binary Mode） and the LDP（Local Directional Pattern） face expression method of feature extraction. The dimension of improved LBP has decreased significantly, more considered the space structure information. The method of LDP has the very strong antinoise ability, and better edge information extraction. Based on the two methods can be complementary advantages, the image block branch area with improved LBP for first feature extraction. Subspace of face use LDP area feature extraction. Finally the features of the improved LBP extraction and extraction of the LDP extraction characteristics connected as the general features. JAFFE database of experiment proves the proposed method has more effectively improving the face expression recognition.

Key words: facial expression recognition, Local Binary Mode（LBP）, Central Binary Mode（CBP）, Kirsch operator, Local Directional Pattern（LDP）

摘要： 提出了一种结合改进的LBP（局部二值模式）和LDP（局部定向模式）的人脸表情特征提取方法。改进的LBP维数明显降低，更多地考虑了空间结构信息且计算速度得到了提高。LDP方法具有很强的抗噪能力，更好地提取边缘信息。基于两种方法可以优势互补，先对图像分块子区域用改进的LBP进行特征提取，再用LDP对脸部子区域进行特征提取，最后把改进的LBP提取的特征和LDP提取的特征按顺序串接起来作为总特征，用最近邻进行分类。在JAFFE表情库进行了实验，证明提出的方法能够有效地提高人脸表情的识别率。

关键词: 表情识别, 局部二值模式, 中心化二值模式, Kirsch算子, 局部定向模式