Multi-Sensor Image Fusion Based on Visual Saliency Detection

Abstract

In military, medical imaging, concealed weapon detection, remote sensing and digital photography applications single image capture may not always sufficient to provide entire information of a target scene. For this purpose more than one image of the same scene has to be captured and useful information from these complementary images has to be merged into a single image. Image fusion is the phenomenon of merging useful source imagery content for better visual understanding of a situation. Here, we introduce a new image fusion method depends on two-scale image decomposition and visual saliency detection for multi-sensor images. Approximation and detail layers are extracted from input images by employing a mean filter. Visual saliencies of input images are computed using visual saliency detection process. Detail layers are fused with help of decision map based on visual saliencies. Approximation layers are averaged to get the final approximation layer. Finally, combined image is obtained by the linear combination of final approximation and detail layers. Proposed method is very beneficial because the visual saliency detection process explored in this paper can highlight image features (visually significant information) very well with full resolution. Therefore, the proposed decision map is able to transfer necessary and complementary data of input imagery into the combined image. In contrast to multi-scale decomposition fusion techniques, our technique is computationally simple since two-scale image decomposition is sufficient to achieve satisfactory results. Proposed technique outcomes are compared with the outcomes of state-of-the-art techniques. Our technique outperforms the existing ones.