Date of Publication :25th June 2024

Abstract:There has been a surge of interest towards deep learning based, image fusion methods in recent years. Through the process of image fusion, complimentary information is extracted from images that have been captured by multiple sensors. Irrelevant characteristics are screened out and the remaining relevant information is combined to enrich the detail and quality of these images. In the context of low light, image fusion techniques face difficulties while preserving the details and diminishing the noise produced in the resulting fused image. This occurs mainly due to the lack of visibility caused by insufficient lighting. Such conditions severely impact the fused images generated by the model. This research paper aims to conduct a comparative analysis of several state-of-the-art, deep learning based image fusion models for low light surveillance applications. Additionally, our paper will investigate the merits and challenges corresponding to each method in the context of low light image fusion. The results of our comparative analysis revealed that ‘SwinFuse’ exhibited superior performance when compared with other methods in preserving image details and reducing noise in the fused images.

Reference :