RGB-D Fusion Based on Fuzzy Optimization for Salient Object Detection

Sudipta Bhuyan, Debashis Sen, Sankha Deb

Abstract: The objective of RGB-D salient object detection is to identify visually distinct objects from both depth and RGB images. The fusion of depth information with RGB has gained significant research attention in this field since salient objects can appear in one or more modalities. This paper proposes a fusion scheme for detecting multiple salient objects from RGB-D data by integrating multiple prior maps using a fuzzy optimization framework. First, We generate several prior maps, which are considered to be fuzzy. A fuzzy divergence measure is then utilized to minimize the discrepancies between the different prior maps during the fusion process, ultimately maximizing the number of detected salient objects in the final fused map. Our proposed framework estimates the optimal fuzzy membership parameters to address the boundary ambiguity between salient regions and non-salient backgrounds. Experimental results on different databases demonstrate the efficiency of the proposed method compared to existing RGB-D salient object detection methods.