Infrared Image Super-Resolution via Lightweight Information Split Network

TL;DR

This work tackles infrared image super-resolution under resource constraints by introducing LISN, a lightweight network built from shallow and deep feature extraction, dense feature fusion, and high-resolution reconstruction modules. The core innovation is the Lightweight Information Split Block (LISB), which fuses channel splitting, shift operations, a Residual Depth-wise Convolution Block, and contrast-aware channel attention to achieve high SR quality with few parameters, complemented by a Sobel-based edge loss for sharper edges. Across infrared benchmarks, LISN achieves state-of-the-art PSNR/SSIM with significantly lower FLOPs and memory usage than competing methods, demonstrating strong practical viability for deployment on constrained infrared devices. The work provides a detailed ablation study to justify design choices (e.g., six LISBs, RDB/CCA contributions) and suggests future integration with infrared small-target detection to broaden applicability.

Abstract

Single image super-resolution (SR) is an established pixel-level vision task aimed at reconstructing a high-resolution image from its degraded low-resolution counterpart. Despite the notable advancements achieved by leveraging deep neural networks for SR, most existing deep learning architectures feature an extensive number of layers, leading to high computational complexity and substantial memory demands. These issues become particularly pronounced in the context of infrared image SR, where infrared devices often have stringent storage and computational constraints. To mitigate these challenges, we introduce a novel, efficient, and precise single infrared image SR model, termed the Lightweight Information Split Network (LISN). The LISN comprises four main components: shallow feature extraction, deep feature extraction, dense feature fusion, and high-resolution infrared image reconstruction. A key innovation within this model is the introduction of the Lightweight Information Split Block (LISB) for deep feature extraction. The LISB employs a sequential process to extract hierarchical features, which are then aggregated based on the relevance of the features under consideration. By integrating channel splitting and shift operations, the LISB successfully strikes an optimal balance between enhanced SR performance and a lightweight framework. Comprehensive experimental evaluations reveal that the proposed LISN achieves superior performance over contemporary state-of-the-art methods in terms of both SR quality and model complexity, affirming its efficacy for practical deployment in resource-constrained infrared imaging applications.

Figures (5)

• Figure 1: The architecture of the LISN for lightweight infrared image SR. LISN consists of four key components: Shallow Feature Extraction (SFE), Deep Feature Extraction (DFE), Deep Feature Fusion (DFF), and Infrared Image Reconstruction (IIR).
• Figure 2: The architecture of the LISB for deep feature extraction. LISB is primarily composed of the following modules: Shift Building Block (SBB), Residual Depth-wise Convolution Block (RDB), Contrast-aware Channel Attention Block (CCA)
• Figure 3: This demonstrates the main components of LISB. (a) The architecture of CCA. (b) The architecture of PA. (c) The architecture of shift block. (d) The architecture of RDB. The DWC3 means that the kernel size is $3 \times 3$ for depth-wise convolution.
• Figure 4: Visualization comparison of different pruning methods on CVC-09-1K dataset.
• Figure 5: The effect of PSNR measure of the proposed model with respect to training epochs for scaling factors of 4 and 2.