A low-complexity learning-based algorithm for joint application of demosaicking and up-scaling

This paper proposes a low-complexity algorithm based on machine learning to deal with demosaicking and upscaling processes simultaneously. First, the color difference planes are calculated to extract the edge orientation and reconstruct the missing channel of G. Then the correlation of neighboring pixels is adopted to reconstruct the missing channels of RB and refine reconstructed the color planes. Finally, a novel machine learning structure is employ to up-scale image by optical interpolation kernels on the basis of designate feature descriptor. The refined operation and machine learning structure are two crucial contribution in this paper, which bring about decrease of the combination error and improvement of reconstructed quality. In experimental result, it is confirmed that the proposed method gets better reconstructed quality with lower cost of calculation than previous related methods.

I. Introduction

Digital images is an important application nowadays, which issues of demomaicking and up-scaling are developed and researched for many years respectively. Typically, image demomaicking implies the transformation from Bayer pattern [1] to full color image on digital camera, as shown in Fig 1. Bayer pattern is also called as CFA image. The usage of color difference plane is applied to demosaicking issue in [2] [3], which is referenced by further related method [4]–[7].