Single Frame Image Super Resolution Using ANFIS Interpolation: An Initial Experiment-Based Approach

Abstract

Image super resolution is a classical problem in image processing. Different from most of the existing super resolution algorithms that work on sufficient training data, in this work, a new super resolution method is proposed to handle the situation where the training data is insufficient by the use of ANFIS (Adaptive Network-based Fuzzy Inference System) interpolation. ANFIS interpolation aims to interpolate an effective ANFIS given only sparse data in the problem area of interest, with the assistance of two well trained ANFISs in the neighbourhood areas. The interpolated ANFIS constructs mappings from low resolution images to high resolution ones, which provides an effective mechanism for further inference of high resolution images from given low resolution ones. Experimental results indicate that the proposed approach entails improved super resolution performance for situations where there is a shortage of training data.