"IMDN-Plus: Enhancing Lightweight Super-Resolution with Hybrid Attentio" by Peiqi Yu

Author

Peiqi Yu

Date of Award

3-18-2025

Document Type

Thesis - SCU Access Only

Publisher

Santa Clara : Santa Clara University, 2025

Degree Name

Master of Science (MS)

Department

Computer Science and Engineering

First Advisor

Nam Ling

Abstract

Super-resolution (SR) is a critical task in computer vision, aiming to reconstruct high-resolution images from low-resolution inputs. While deep learning-based SR models have achieved impressive results, many are computationally expensive and unsuitable for realtime applications. IMDN (Information Multi-Distillation Network) is a lightweight SR model designed for efficiency, but it has limitations in spatial attention, hierarchical feature learning, and edge preservation.

In this work, we propose IMDN-Plus, an enhanced lightweight SR model that integrates Hybrid Attention (CBAM + Swin Transformer) to refine feature extraction, Deeper Feature Distillation to improve texture recovery, and Charbonnier Loss to enhance edge sharpness. These modifications allow IMDN-Plus to achieve better perceptual quality without significantly increasing computational cost.

We evaluate IMDN-Plus on the DIV2K validation set, where it achieves an average PSNR improvement of +0.9 dB and an SSIM gain of +0.0158 over IMDN. Qualitative results further demonstrate sharper textures and reduced artifacts in high-frequency regions. These findings suggest that IMDN-Plus is a practical solution for real-time SR applications, balancing efficiency and image quality.

Share

COinS