Multi-feature aggregation attention for efficient image super-resolution

Xiangyuan Zhu

doi:10.1007/s00371-025-04304-x

Multi-feature aggregation attention for efficient image super-resolution

2025-12-19

0

PRE

AI

DOI：10.1007/s00371-025-04304-x

Original

Original request for help

Save

Abstract

En 中文

Image super-resolution aims to enhance low-resolution images by reconstructing high-resolution details. Despite significant advancements, existing methods often struggle with computational efficiency and model complexity. This paper introduces a novel efficient SR method based on a multi-feature aggregation attention mechanism. We propose a cross-feature attention module to refine feature interactions and a multi-feature aggregation attention module to redistribute feature weights. Our approach achieves state-of-the-art performance on benchmark datasets, significantly improving image clarity with fewer parameters. Here, we show a PSNR improvement of up to 0.41dB on the Set5 dataset at a 4x scaling factor, demonstrating the effectiveness of our method in balancing performance and computational efficiency. The source code can be downloaded at https://github.com/zxycs/MFAASR.

Keywords:

Feature aggregation

Attention mechanism

Super-resolution

Efficient strategy

Multi-feature aggregation attention for efficient image super-resolution

Abstract

Journal

Organization