A Block-matching Image Registration Algorithm for Video Super-resolution Reconstruction
一种用于视频超分辨率重建的块匹配图像配准方法

Image registration is one of the key components in super-resolution reconstruction, and it directly affects the quality of the reconstructed image. On the basis of adaptive rood pattern search (ARPS) block-matching algorithm, and according to the correlation between various images in wavelet domain, an hierarchical block-matching algorithm --- improved adaptive rood pattern search algorithm based on wavelet transform (W-IARPS) is proposed. Searching matched-macroblocks in wavelet domain can effectively reduce the number of search points, and the ratio of peak signal to noise of registered images decreases less than 0.l dB compared to the exhaustive-search algorithm. It means that the high registration accuracy is guaranteed. Finally, the projections onto convex sets (POCS) method is used to reconstruct the super-resolution image from the registered images, and it can achieve better visual effects. The simulation results show that the algorithm has a high registration accuracy and reliable reconstruction results.