Abstract: A super resolution (SR) reconstruction algorithm of video sequence based on inter-frame homography transform is presented in this paper. The mathematical model of SR is introduced using the least square principle. Hierarchical model-based motion estimation is adopted to compute the adjacent homography matrix. For non-adjacent frames, the homography matrix is estimated with guided feature registration based on RANSAC algorithm, which gives sub-pixel accuracy. Finally, a conjugate gradient algorithm is used to reconstruct the SR frames, and computation of matrix-vector multiplication is studied in detail. Experimental results demonstrate effectiveness the proposed method.

Key words: video processing, super resolution reconstruction, homography transform, RANSAC algorithm, CG algorithm