Super resolution using block-matching motion estimation with rotation

Abstract

Objects in real-life recorded image sequences exhibit rotation along with linear motion. This should be taken into consideration when doing super-resolution work on video sequences. The work presented here involves determination of motion-vector field that includes rotation of blocks in the registration step which is the most important step of the super-resolution restoration work. Results of proposed approach are compared against the results that uses exhaustive block search that does not consider rotation. Compared approaches are applied upon low resolution sequences generated from their high resolution counterparts via downsampling. Comparison is done over the peak-signal-to-noise-ratios (psnr) of the output images with improved resolution. Visual quality assessment is also performed.