LSKN-MFIF: Large selective kernel network for multi-focus image fusion

Multi-focus image fusion (MFIF) is an image enhancement technique that compensates for the limited depth of field of optical lenses. It has broad application prospects across multiple disciplines such as digital photography, biomedical imaging, and machine vision, similar to other advanced visual tasks. This study proposes a large selective kernel network, referred to as LSKN-MFIF, which is designed to improve multi-focus image fusion. This network has the capability to dynamically adjust its extensive spatial receptive field, effectively expanding the receptive field and capturing multi-scale global information, fully extracting and integrating multi-scale context to better identify various focused regions in the image. More specifically, LSKN-MFIF extracts multi-scale global features through a large selective kernel module (LKSB) that decomposes large kernel convolutions, utilizes spatial feature selection for feature aggregation, and then enhances representation capability by extracting local information through a gated differential convolution block (GDCB), thereby generating accurate decision maps. Experimental results show that the proposed methodology surpasses current multi-focus image fusion techniques in terms of both subjective visual quality and objective performance metrics.