Image face calibration positioning based on 3000FPS algorithm

Image face calibration positioning based on 3000FPS algorithm

In the face recognition tasks, face pose can have influence on the results of recognition. To improve the accuracy of recognition, we introduced 3000FPS algorithm for face image calibration positioning. Random forests is used to establish each key point, and the output of every tree in random forests, binary feature, is exploited to compose local binary feature, which is used to train a softmax classifier. The experiment show that the recognition accuracy of the algorithm can reach 93.33% on the Helen face dataset, and 94.14% on the LFPW. Meanwhile, multi-face can be aligned and located even though the input image is partially obscured.