Blind reverse CFA demosaicking for the reduction of colour artifacts from demosaicked images

摘要

When a single image sensor is used to capture a colour image, only one colour at each pixel location is acquired. Demosaicking is required to estimate the other missing colour information in order to produce a full colour image. Digital images taken by early generation digital cameras using primitive demosaicking algorithms have resulted in inferior quality images with colour artifacts. In general, it is difficult to apply filtering for the removal of those artifacts. There has been active research in the development of demosaicking algorithms for digital cameras resulting in higher quality demosaicked images with minimal colour artifacts. If those inferior quality images could be re-demosaicked by making use of the advance in demosaicking algorithms, a better output image can be produced without filtering. In this paper, we proposed a novel technique to firstly extract the true sensor pixel values using blind reverse demosaicking from the inferior colour image input. We then apply a state-of-the-art demosaicking method to reproduce the output image with minimal colour artifacts. One advantage of our proposed blind reverse demosaicking method is that an image not in RAW format (digital negative) can still be improved in quality with the advancement of demosaicking algorithms.