Stereo Image Super Resolution Reinforced by Non-Local Means Denoising (NLMD) Algorithm

For single-image reconstruction and enhancement techniques, deep convolutional neural networks (CNNs) performed better at producing super-resolution (SR) images from the details of low-resolution (LR) images. As a result, by utilizing intra- and cross-view information, Stereo Super Resolution approaches increased the effectiveness of these convolutional networks in recovering the original texture of individual images and provided further information about their tiny details. This paper presents a model of the Stereo Image Super Resolution Algorithm (SSR<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">nlmd</inf>) reinforced by non-local means denoising (NLMD). Using two weight-sharing networks (NAFNet Blocks) to extract intra-view features of the left and right input image pair, then fuse these features to SR reconstruction, is one method for recovering super-resolution (SR) image details from low-resolution and improving the visual rendering of their surface textures. The resolution levels of the images generated by NAFNet Blocks have been reinforced using the non-local means denoising algorithm to remove noise and distortion and obtain an explicit scene close to the original image (HR Image). Quantitative results of (SSR<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">nlmd</inf>) model demonstrated that our method outperformed state-of-the-art methods in terms of metric (PNSR/ SSIM), It was 31.29dB/0.8722 of 0376 Image on the DIV2K dataset. Quantitative scores on the BSD100 dataset were 38.88dB/0.9739 of 3096 Image. The result on the Set5 dataset was 32.29dB/0.8879 for the baby Image. The scores on the Set14 dataset were 21.46dB/0.5414 for the baboon Image. The visual evaluation of the images reconstructed by our method achieved high accuracy and clarity in the scene with less noise and enhanced quality.