Predictive Filter Flow Network for Universal Demosaicking

Abstract

Demosaicking is an image reconstruction process for restoring full-color images from color filter array (CFA) data. In recent years, many deep convolutional neural network (CNN)-based demosaicking methods have been reported, and state-of-the-art accuracy has been achieved. In this paper, we propose a novel demosaicking method using the predictive filter flow (PFF) network for various CFA patterns. The PFF is a model that predicts a spatial variant linear filter that transforms an input image into a target image. To incorporate the PFF into demosaicking, the proposed network synthesizes the filter flow corresponding to each channel by means of a network trained by integrating RGB channels. Our model, designed to apply demosaicking with the PFF to various CFA patterns, provides versatility and extensibility. Experimental results demonstrate that the proposed method provides better or competitive results compared with several state-of-the-art deep-CNN-based demosaicking algorithms.