An efficient framework for image interpolation using weighted surface approximation

Although it has been recognized that different textual contents in an image need to be treated differently during accurate image interpolation, how to classify these contents well has been a difficult problem due to the inherent complexity in natural images. In this paper we propose an efficient image interpolation framework with a novel weighted surface approximation approach. The key is that the weighted mean squared error of the approximation can be converted to a continuously distributed probability of a pixel belonging to a local smooth region or a textural one, thus essentially making a soft pixel classification. In addition, the fitted local surface provides an estimate of the pixel value under the smooth region assumption. This estimate is then fused with the estimate from the texture region assumption using the previously obtained probability to yield the final estimate. Experimental results show that the proposed framework consistently improves over typical state-of-the-art methods in terms of interpolation accuracy while maintaining comparable computational complexity.