MULTI-NOISENET CNN MODEL FOR COLOUR IMAGE RESTORATION

Abstract

Artificial intelligence and deep learning have led to a rise in the use of this technology in several fields, including image noise reduction. If you want to further process the image for purposes like object segmentation, detection, tracking, etc. then you need to eliminate the noise from the image and restore a high quality image. But there are substantial discrepancies between the numerous deep learning approaches for picture denoising. In particular, deep learning-based discriminative learning can efficiently address the problem of Gaussian noise. Pretrained Convolutional Neural Network (CNN) model for picture denoising is presented in this paper. An advantage of using this CNN model over traditional image denoising methods like Wiener and median filtering is that it can be fine-tuned during the filtering process, whereas in old-style denoising, the limits of these algorithms are secure and cannot be changed, which is called lack of adaptivity. This study uses four different datasets and a variety of performance indicators to demonstrate the general five tasks of picture denoising. This suggested model (Multi-noiseNet) directly estimates the latent clean picture and removes noise by utilising a residual learning technique. Data from Set14 dataset showed that the suggested model had a peak signal-to-noise ratio (PSNR) of 41.17 percent and an structural similarity index measure (SSIM) of 0.972 when salt and pepper noise was 10%.