Single image resolution enhancement by patch based learning

In this paper we propose a resolution enhancement approach using patch based learning method. The patches in natural image are occurred repeatedly many time inside the image. This tendency of occurring similar patches is within the same scale, as well as across different scales. Hence the present method estimates a function to predict the pixels of a High Resolution patch using its corresponding Low Resolution pixel and their spatial neighbourhood. This method gives best results in terms of reconstruction accuracy of output High Resolution patch size when given input Low Resolution patch size is increased. The patch based single image super-resolution is more efficient than earlier bilinear and bicubic methods. The present method uses zero component analysis which is used as preprocessing step for improving perceptual element such as edges.