Image Registration using Combination of GPOF and Gradient Method for Image Super Resolution

Abstract

Super Resolution implementation using multi-frame super resolution has been an expensive topic in the literature. Multiframe Super-Resolution is to generate the high-resolution image from multiple low-resolution images perspectives of a same scene. Most important part of multi-frame Superresolution is Image Registration; that estimates the translation, rotation and scaling parameters and also aligns images. In this paper, they propose a combination of Gaussian Pyramid Optical Flow (GPOF) Registration method and Gradient method for constructing Super Resolution image. In the proposed approach, they focus on the movement model of the image registration GPOF, which reach the sub-pixel and allows for the large pixel motions, while keeping the size of image neighborhood relatively small. And apply Gradient method, which can accurately perform precise registration with the amount of image movement is small between the two images; it can get the one reconstructed image. They get the better results compares with the others registration methods. And lastly, they apply Discrete Wavelet Transform (DWT) image interpolation algorithm; they can get the high resolution image. Experiment results show that the HR image by their proposed method have much higher quality than other methods.