Optimized CNN Based Demosaicing and Empirical Wavelet Transform for Denoising in Colour Images

For modern digital cameras to capture realistic images, several image processing procedures must be carried out in order. The first two processes are often associated with denoising and demosaicing, where the former seeks to minimize sensor noise and the latter produces color pictures from a sequence of light intensity values. Modern techniques aim to jointly address these challenges, i.e., joint denoising-demosaicing, which is an ill-posed issue by nature as noise modulates two-thirds of the brightness data. An extensive analysis of using convolutional neural network (CNN) to solve the demosaicing and denoising problem is presented in this publication. Since the CNN model is so adaptable, it may be used with any CFA architecture for demosaicing. Images may be converted between different color models using color space conversion transforms. To manipulate and alter color, an RGB (Red, Green, and Blue) picture can be transformed to the HSL or HSV color models. Using three distinct CFAs, we train CNN models for demosaicing and outperform current techniques in the process. Tests demonstrate that the automatically generated demosaicing technique and the automatically identified CFA pattern work better together than they do alone.