<p>Infrared and visible image fusion aims to integrate the thermal radiation information from infrared images with the detailed texture information from visible images into a single image to enhance understanding of various scenarios. Existing multiscale decomposition algorithms often suffer from the loss of fine texture details and contrast. To overcome this limitation and enhance texture details and contrast, we propose a novel method for infrared and visible image fusion based on multiscale Gaussian total variation (MGTV) and an adaptive entropy and structural similarity index-weighted fusion strategy. First, the source images are decomposed into high-, medium-, and low-frequency layers using MGTV decomposition. For the medium-frequency layer, an adaptive fusion rule based on local entropy and structural similarity index is applied to preserve key structural details. The low and high-frequency layers are fused using the maximum selection strategy and weighted least-squares fusion strategy, respectively, to ensure edge sharpness and contrast enhancement. Finally, the fused image is obtained by an adaptive strategy that joins all layers. Experimental results on the TNO and <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="371_2025_3840_Article_IEq1.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="56" /> </InlineMediaObject> <EquationSource Format="TEX">$M^3FD$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <msup> <mi>M</mi> <mn>3</mn> </msup> <mi>F</mi> <mi>D</mi> </mrow> </math></EquationSource> </InlineEquation> datasets demonstrate that the proposed method achieves significant improvements, showing notable enhancements in EN, SF, SD and VIF compared to existing methods. Specifically, the proposed method achieves up to 15% improvement in edge sharpness and contrast preservation compared to traditional methods. In particular, the method demonstrates a notable increase in edge sharpness, with improvements in structural edge contrast by approximately 18%. Additionally, texture details are preserved with a 14% improvement, and contrast enhancement is improved by 12%, as quantified by the VIF index. These improvements lead to clearer and more detailed fused images. Both qualitative and quantitative evaluations confirm the superior fusion performance of the proposed method, which meets well with human visual perception. The code is available at <a href="https://github.com/lh-ite/MGTV_Fusion">https://github.com/lh-ite/MGTV_Fusion</a>.</p>

Enhancing infrared and visible image fusion through multiscale Gaussian total variation and adaptive local entropy

期刊论文
发表日期：

2025年3月11日

查看全文

Hao Li,
Shengkun Wu,
Lei Deng,
Chenhua Liu,
Yifan Chen,
Hanrui Chen,
Heng Yu,
Mingli Dong,
Lianqing Zhu

摘要

Infrared and visible image fusion aims to integrate the thermal radiation information from infrared images with the detailed texture information from visible images into a single image to enhance understanding of various scenarios. Existing multiscale decomposition algorithms often suffer from the loss of fine texture details and contrast. To overcome this limitation and enhance texture details and contrast, we propose a novel method for infrared and visible image fusion based on multiscale Gaussian total variation (MGTV) and an adaptive entropy and structural similarity index-weighted fusion strategy. First, the source images are decomposed into high-, medium-, and low-frequency layers using MGTV decomposition. For the medium-frequency layer, an adaptive fusion rule based on local entropy and structural similarity index is applied to preserve key structural details. The low and high-frequency layers are fused using the maximum selection strategy and weighted least-squares fusion strategy, respectively, to ensure edge sharpness and contrast enhancement. Finally, the fused image is obtained by an adaptive strategy that joins all layers. Experimental results on the TNO and $M^3FD$ $M^{3} F D$ datasets demonstrate that the proposed method achieves significant improvements, showing notable enhancements in EN, SF, SD and VIF compared to existing methods. Specifically, the proposed method achieves up to 15% improvement in edge sharpness and contrast preservation compared to traditional methods. In particular, the method demonstrates a notable increase in edge sharpness, with improvements in structural edge contrast by approximately 18%. Additionally, texture details are preserved with a 14% improvement, and contrast enhancement is improved by 12%, as quantified by the VIF index. These improvements lead to clearer and more detailed fused images. Both qualitative and quantitative evaluations confirm the superior fusion performance of the proposed method, which meets well with human visual perception. The code is available at https://github.com/lh-ite/MGTV_Fusion.