Adaptive Multiscale Wavelet and Deep Structured Architecture for Generalized Edge-Preserving Image-Denoising Framework
Srinivasa Rao Thamanam, K. Manjunathachari, K. Satya Prasad · International Journal of Image and Graphics · 2025
Image denoising is the procedure of eliminating noise from an image. One of the recurring issues in the field of image analysis and processing has always been noise reduction. However, it is crucial to maintain the edges of an image while lowering the noise levels. Therefore, it is preferable to keep crucial elements like corners, edges, and other sharp structures undamaged while denoising. Although many techniques and algorithms have been developed in the existing works, those techniques still have drawbacks in the quality of images. Those methods mostly lose something while decreasing noise, or they fail to reduce noise sufficiently. Moreover, they could not handle the minor structures and area boundaries effectively. Therefore, modern techniques are needed to decrease the image noise without damaging the edges. To address these deficiencies, we developed an effectual deep learning-aided edge-preserving image-denoising framework. The visibility of essential features in images is enhanced by eliminating the noise in images. At first, the mandatory images are accumulated from various sources of databases. Subsequently, the collected images are given as the input into the Adaptive Multiscale Wavelet (AMW) for the image decomposition process. This approach has the capability to handle large databases. Here, the parameters are tuned using the Modified Manta Ray Foraging Optimization Algorithm (MMFOA) to improve the performance of image decomposition. Then, the image denoising is performed using the Residual Attention Network (RAN). After denoising, the Inverse AMW (IAMW) is performed to reconstruct the denoised images. Finally, the effectiveness of the developed model is compared and contrasted with numerous baseline models to prove its superiority over others.