关键词: 表情识别, 部分遮挡, 对称变换, 积分投影

Abstract: In order to solve the problem that six basic facial expressions are difficult to recognize when the human eyes are partially occluded due to the fringe of hair or hats，a new method based on symmetry transformation is presented.Referring to the facial geometrical properties，a vertical integral projection arithmetic is adopted on normalization image.Through the midline detection algorithm which is used to judge how much the eyes are occluded，the midline of the face can be got.When the occlusion is intolerant，the image has to be renovated by symmetry transformation.Experimental results indicate that better recognition effect can be got on the same conditions of the feature extraction measure and classifier by this method.What’s more，it can tolerate the head’s deflection within a certain range.

Key words: facial expression recognition, partial occlusion, symmetry transformation, integral projection