Reconstruction of a HR image by simultaneous POCS and regularized block matching

摘要

In the reconstruction of high resolution (HR) images from low resolution (LR) image frames, the accuracy in the inter-frame motion estimation is a critical factor. With a motivation that motion estimated over the reconstructed HR frames would have higher accuracy, several attempts have been made to perform motion estimation and HR reconstruction simultaneously. In this paper, we propose a HR reconstruction method, in which a regularized block matching for motion estimation and POCS-based reconstruction are performed in turns. In experimental results, regularized block matching produces reliable motion estimation since the motion of an image pixel is highly correlated with the motion in neighboring region. The improved accuracy of the motion estimation results in higher PSNR of the reconstructed HR images.