Research On Facial Expression Recognition Based On HOG

Facial expression recognition draws more and more attention of researchers and it has gradually become a hot topic in the field of computer vision and pattern recognition. In recent years, the facial expression recognition technology has achieved significant development, however, there are still some key problems need to be solved urgently, such as the low recognition rate of the algorithm, the week robustness of the algorithm to noise, and the long running time of the algorithms.To solve these problems above, it mainly makes some research on feature extraction and the main research contents and innovative work are as follows.1. It adopts Histograms of Oriented Gradients(HOG) to extract the feature of an expression image, and it studies the principle of HOG. What’s more, it takes spacial arrangement between local features of the expression image into consideration and it proposes a new facial expression recognition algorithm based on Pyramid of Histograms of Oriented Gradients. The experimental results show that the new expression feature reduces 2/3 in its dimension compared with the traditional HOG feature extracted from a whole expression image and it can express the shape information of the expression image and the special arrangement between local features much better. What’s more, the efficiency and recognition rate of the new algorithm make more obvious improvement.2. It proposes a new anisotropic diffusion algorithm based on human visual system, which uses luminance difference and the gradients to adjust the diffusion coefficient. The new method can retain the weak edge information and weak details much better while filtering out the noise. The experimental results on JAFFE database show that the new facial expression recognition combining the improved anisotropic diffusion algorithm and HOG is an effective algorithm and it is robust to noise.3. It studies the attention mechanism in human visual system and it marks the key points of the face according to Active Shape Model(ASM) algorithm. Then it extracts the local HOG feature of the key points. In addition, it studies the principle of multiscale analysis and it obtains multiscale HOG feature, which eliminates the interference of redundant information and reduces the feature dimension. The experiment results show that the new facial expression recognition algorithm based on visual attention mechanism and multiscale HOG is much more effective and it can describe the facial expression information much better.