SAMF: Small-Area-Aware Multi-focus Image Fusion for Object Detection

TL;DR

This work tackles the challenge of preserving small, uncertain, and boundary-focused regions in multi-focus image fusion to aid object detection. It introduces SAMF, a small-area-aware MFIF framework that combines an Enhanced Pre-fused Image Acquisition stage with a multiscale high-frequency fusion, producing an Enhanced Pre-fused Image from PF and FH, where EPF = PF + FH. A novel three-region segmentation strategy integrates two-region decisions with a three-region decision map to classify pixels as focused, defocused, or uncertain, guiding fusion through a final FMP decision rule. Extensive experiments on datasets including Road-MF, Lytro, and MFI-WHU show SAMF outperforms nine baselines in both objective metrics, $Q_{MI}$, $Q_M$, $Q_S$, $Q_{CV}$, and practical object-detection performance, demonstrating its potential for robust, small-area-aware fusion in driving and surveillance contexts; future work will address disturbances from misalignment and varying resolutions.

Abstract

Existing multi-focus image fusion (MFIF) methods often fail to preserve the uncertain transition region and detect small focus areas within large defocused regions accurately. To address this issue, this study proposes a new small-area-aware MFIF algorithm for enhancing object detection capability. First, we enhance the pixel attributes within the small focus and boundary regions, which are subsequently combined with visual saliency detection to obtain the pre-fusion results used to discriminate the distribution of focused pixels. To accurately ensure pixel focus, we consider the source image as a combination of focused, defocused, and uncertain regions and propose a three-region segmentation strategy. Finally, we design an effective pixel selection rule to generate segmentation decision maps and obtain the final fusion results. Experiments demonstrated that the proposed method can accurately detect small and smooth focus areas while improving object detection performance, outperforming existing methods in both subjective and objective evaluations. The source code is available at https://github.com/ixilai/SAMF.

Figures (4)

• Figure 1: The application of MFIF in automatic driving.
• Figure 2: Flowchart of the proposed algorithm.
• Figure 3: Fusion results of different methods on Road-MF dataset.
• Figure 4: Fusion results of different methods on Lytro and MFI-WHU datasets.