Occlusion face recognition based on partition selection and gabor wavelet

Partial occlusion face recognition is a difficult problem in face recognition applications. Since the occlusion parts have a weak contribution to face recognition, these parts should be excluded when classifying. To solve this problem, an occlusion face recognition method combining partition selection and Gabor wavelet is proposed. Firstly, the image is divided into unconnected sub-blocks and the occlusion area in the face image is determined according to the image root mean square error information. Secondly, Gabor filters with 5 scales and 8 directions are used to extract features from unoccluded partitioned images. Then cosine similarity is used as texture separator to identify and classify the extracted features. Finally, the recognition results of the unoccluded partition of the test image are fused for decision making, and the recognition accuracy and other rating indicators are counted. The performance of the algorithm is tested in the data set containing different occlusions, and the recognition accuracy rates reach more than 95%.