To resolve the problems existed in the estimation of motion in video sequences, a novel method combining the gray projection in a spatial domain and the phase correlation in a frequency domain was proposed. Firstly, the gray projection algorithm was adopted to coarsely register images to entire pixel accuracy, which calculates the gray correlation function for the spatial domain in row and column orientations and obtains the pixel-level motion vector between two sequential images by comparing correlation characters. Then, the phase correlation algorithm was adopted to refinedly register images to sub-pixel accuracy, which uses power spectrum information of images to decrease the image dependence. Furthermore, the fitting method of paraboloid surface based on least-square was used to fulfill the estimation of image displacement parameters. To verify the algorithm experimentally, some samples were simulated. The results show that this method can detect the displacement parameters accurately and efficiently, and can offer the image displacement in a accuracy of 0. 01 pixel and the maximum registration error less than 0. 004 8 pixel.
国家科技期刊平台
NSTL
万方数据
维普
