Implement And Optimization Of Face Recognition Algorithm On Dual Dsp

Face recognition is very difficult in pattern recognition, but it also has a very important theoretical value and practical application in the field of pattern recognition. It has great prospects in the authentication, human-machine interaction and visual surveillance application field.This paper mainly talked about the implementation and optimization of Face Recognition System in ADSP-BF561. Face recognition was researched and developed according the image preprocessing, face detection, facial feature extraction and face comparison.In the face detection stage, firstly, using the elliptical skin color model, the entire image binarization and noise remove to crude locate the face region, then using integral projection method to locate the eyes, then taking the eyes of the distance as a standard, locating the face according to the characteristics of the face. Finally, making image size normalization to get the standard face image.In the feature extraction, this paper analysis the advantages and disadvantages of DCT and PCA. Integration of feature extraction methods of DCT and PCA was used. First, it useed the DCT transform to extract DCT feature vector as PCA input to get the final feature vector. Finally it used the markovian distance to do the compare treatment.At the end of this paper, it studied the optimization strategy of the face recognition in the BF561, such as compiler optimization, memory optimization, dual-core optimization, code optimization and so on. It basically reached the effect of real-time processing.The results of the experimental show that the recognition rate of the method proposed in this paper can reach 95% in the case of a single human face and the front face. The result shows that the method is practical.