Research And Implementation Of Face Detection Algorithm Based On DSP

Face detection technology is a very popular research topic. At present, face detection technology is relatively mature with controlled lighting conditions and the users’cooperating.With extensive research to some fundamental methods of face detection, face recognition and image matching, this paper uses the method that bases on the color of the skin to realize face detection and the SIFT algorithm to realize target face recognition. The SIFT algorithm has a strong matching ability and robustness. And it can be applied to face recognition, image registration, image retrieval and so on. With matching images between the SIFT feature vector of the candidate faces and the target face, this paper achieves the target face recognition fast and accurately.After determining the overall system program, this paper uses MATLAB and DSP to simulation and implementation.On the MATLAB platform, this paper completes the extraction of color range of skin, the location of face region and the extraction of the SIFT feature vectors, finally realizes the matching between the target faces and the detected face.In DSP hardware platform design process, firstly, the skin color model, which is counted by MATLAB, is used to determine the candidate skin regions of images; secondly, according to the shape features of people’s faces, the non-face area is excluded, and this can achieve the precise positioning of the faces in the images. In the process of determining the face candidate region, because of the restrictions of DSP memory and DSP computation rate, the method of connected domain of the regional growth is used to determine the face candidate block. With the combination of the structural characteristics of faces, the module, which can exclude interference of pure color blocks, is designed, and substitutes the methods such as edge detection. So, this method can reduce the amount of DSP operations. Finally, image matching of face normalized is completed.The software platform is MATLAB2010, Microsoft visual C++6.0and CCS3.3. The hardware development platform is ICETEK-DM6437-B evaluation module, and its core CPU is TMS320DM6437.The experiment showed that the detection rate of the algorithm in this paper is beyond90%and has a certain degree of stability and a strong practicality.