PROJECT TITLE :

Learning joint demosaicing and denoising based on Sequential energy minimization - 2016

ABSTRACT:

Demosaicing is a crucial first step for color image acquisition. For sensible reasons, demosaicing algorithms must be each efficient and yield high quality ends up in the presence of noise. The demosaicing downside poses several challenges, e.g. zippering and false color artifacts along with edge blur. In this work, we have a tendency to introduce a completely unique learning primarily based methodology that may overcome these challenges. We tend to formulate demosaicing as a picture restoration downside and propose to be told economical regularization galvanized by a variational energy minimization framework that may be trained for various sensor layouts. Our algorithm performs joint demosaicing and denoising in close relation to the $64000 physical mosaicing process on a camera sensor. This can be achieved by learning a sequence of energy minimization problems composed of a group of RGB filters and corresponding activation functions. We evaluate our algorithm on the Microsoft Demosaicing data set in terms of peak signal to noise ratio (PSNR) and structured similarity index (SSIM). Our algorithm is highly efficient both in image quality and run time. We tend to achieve an improvement of up to two.six dB over recent state-of-the-art algorithms.