Face Detection Based on Multi-information Fusion
|School||Southwest Jiaotong University|
|Course||Signal and Information Processing|
|Keywords||face detection feature detection feature location super-resolution fusion Gaborwavelet|
Face detection and facial feature detection are the hotspot in the field of pattern recognition. As a key link of the automatic face recognition system, face detection has got attracted attention gradually. With the expansion of the range of its application, the application background of the face detection has far exceeded the scope of face recognition system. It plays an important role in content-based retrieval, photography, digital video processing, video detection, etc. Although the face detection research has made some achievements at present, face detection in video is still facing great challenges because of the complex environment of the face video.Firstly, the AFS-RS method is adopted to improve the resolution of face images in video. As face images in video often have some problems such as low-resolution, motion blur, less characteristic information and low SNR, the AFS-RS is applied to solve these problems, and get some improvements. The results show that the method can improve the real-time and effectiveness of the face detection in the video.Secondly, a method of face detection based on multi-information is proposed, and the abundant temporal information, color information and gray information in video are fully utilized. The experiment results show that the method of face detection based on multi-information fusion can improve the real-time and the effectiveness of the face detection in video.Thirdly, a method of feature detection and localization based on multiple-parameters Gabor is analyzed. As the Gabor has different parameters, and the different parameters have different representation ability for the different region, we adopt the method to get the rough positioning of the facial feature, and then get the precise position of the eyes and mouth by combine with ASEF and integral projection, respectively. The results show that we can get more precise localization by the above method.Lastly, all the work and the main content are summarized, and something need to improve is analyzed. The prospects are also made for the future work.