Abstract: Firstly, Directshow technology used in the video image to detect moving targets, and on this basis, the video image segmentation and recognition,and finally achieve the image recognition and positioning of the head and the body.Video image processing and recognition using finite difference method and the SIFT characteristics of the pattern mat ching body and head mainly to complete the identification and location. SII门气algorithm on the CUDA-accelerated can get better recognition speed and recognition accuracy by experimental validation, and solve some technical problems of image rccogmtion.

Key words: SIFT，Pattern matching, Image recognition, Scale, Feature