Utilizing convolutional neural network and gray wolf optimization for image super-resolution

Image Super-Resolution (ISR) is a complex task that involves the development of high-resolution (HR) images from low-resolution (LR) inputs, posing a fascinating challenge in the realm of image processing. While deep learning models have shown promise in ISR, the presence of artifacts in images generated by these models often necessitates subsequent post-processing for refinement. This study introduces an innovative approach that combines Convolutional Neural Network (CNN) with Gray Wolf Optimization (GWO) to tackle the obstacles encountered in ISR. The proposed model employs a CNN model for the initial estimation of the upscaled image and incorporates a secondary CNN model utilizing dense layers and hybrid pooling to segment the image and identify regions of uniformity. Simultaneously processing information from the segmented image and the magnification approximation matrix using a GWO-based strategy mitigates the detrimental impact of artifacts on the enlarged image. The GWO algorithm is utilized to dynamically adjust the color layer brightness of individual pixels in distinct regions, tailoring the enhancement process to the specific structural characteristics of each texture region. Performance evaluation of the proposed approach on the Set5, Set14, and Urban100 datasets demonstrates its superiority over existing techniques, yielding enhancements in peak-to-signal noise ratio (PSNR) and structural similarity index measure (SSIM) metrics by a minimum of 1% and 0.5%, respectively.