Face Detection Algorithm Based on Multi-orientation Gabor Filters and Feature Fusion

In order to enhance the accuracy of multi-pose and multi-expression face detection, this paper proposes an algorithm based on multi-orientation Gabor feature fusion of mean and variance of subimages. Firstly, to remove the huge background regions, we segmented images based on YCbCr space and then used two-eye templates to locate faces in skin-color regions by template matching. Secondly, the features were extracted from the face regions which had been removed the nose parts. After filtering by four-direction Gabor filters, the images were divided and then the corresponding constituent parts were formed. Thirdly, we calculated the mean and variance of the constituent parts block by block and processed them with features fusion. At last Support Vector Machine (SVM) and Back Propagation net (BP) were used for classification detection. The experimental results show that the algorithm has higher detection accuracy than other similar algorithms in multi-pose and multi-expression facial detections.