MdedFusion: A multi-level detail enhancement decomposition method for infrared and visible image fusion

Abstract

Listen

Translate article

To improve the infrared target features and preserve the texture details from source images, a simple and efficient multi-level detail enhancement decomposition method based on weighted least squares (WLS) and local statistical edge model (LSEM) is proposed, termed as MdedFusion. First, visibility enhancement of visible image is performed by using guided filter to avoid the effects of low-light environments. Then, infrared image and enhanced visible image are decomposed into a series of smooth parts and detail parts by WLS filter based multi-level decomposition mechanism. Next, in order to highlight edge features, a novel fusion strategy based on LSEM is proposed to make full use of variance difference for fusion of detail parts, and smooth parts are simply fused by weighted average (WAVG) strategy which is an efficient operation in our fusion framework. Finally, the fused image is obtained by reconstructing the fused smooth part and detail part. Specially, particle swarm optimization (PSO) algorithm is introduced to adaptively optimize the filter parameters in our framework. Compared with 16 state-of-the-art fusion methods, the experimental results show that the superiority of our proposed MdedFusion surpass the compared methods in highlighting infrared features and preserving texture detail information. • The weighted least squares (WLS) filter is a classic edge-preserving filter, which can effectively extract background information and texture details from source images to different spatial scales. Therefore, a new multi-level decomposition framework based on the WLS filter is proposed, which ensures that the important information in source images can be fully extracted and retained, and further improves the quality of the final fused images. • A novel fusion strategy based on local statistical edge model (LSEM) is proposed for the fusion of detail layers, in which the variance values of different corresponding image blocks are calculated by the sub-window variance filter (SVF), and then compared and selected. Not only does it achieve adaptive edge feature enhancement for images with different scale detail layers, but also filters out noise interference and redundant information from source images. • We innovatively introduce a particle swarm optimization (PSO) algorithm to optimize the filter parameters in our fusion model. This is the first time the image evaluation metric is designed as a fitness function and applied for iterative optimization, so that the performance effects of different metrics can be evaluated in the fused image. Finally, it can be seen from the experimental results that our MdedFusion has better fusion performance in highlighting infrared target features and texture details.