Image Upscaling and Fuzzy ARTMAP Neural Network

Abstract

In Image upscaling, a higher resolution image is processed from an available low resolution image. There are numerous scenarios and applications where image upscaling is required and performed. The paper proposes a simple image scaling algorithm and implements it with a fuzzy variant of the supervised ART neural network, ARTMAP. The method has been compared with nearest neighbor interpolation, bilinear interpolation and bicubic interpolation, both objectively and subjectively. The present paper is an attempt at understanding the Adaptive Resonance Theory networks while proposing a simple linear upscaling method.