Super-resolution images by support vector regression on edge pixels
Tsz-Chun Ho, Bing Zeng · 2007
Support vector machine (SVM) is a statistical learning algorithm that is capable of estimating high-dimensional functions. Recently, support vector regression (SVR) - the use of SVM for regression - has been used to generate super-resolution images [1]. In this paper, we propose to apply the SVR algorithm on edge pixels only so as to reduce the emboss effect that would appear in the edge region of an enlarged image if the SVR is applied on the entire input image. Such a modification is naturally motivated by the principle that human perception is mainly focusing on edge regions. Doing so can also reduce the overall time consumed during the training process and makes the enlarged image looking more pleasant perceptually.