An Illumination Invariant Approach to Salient Object Detection in Hyperspectral Images

Salient Object Detection (SOD) in Hyperspectral Images (HSI) has traditionally relied on low-level features, with limited exploration of high-level features. Low-level features lack semantic context, making them ineffective for detecting complex or subtle salient objects. Moreover, the inherent sensitivity of HSI to uneven illumination poses additional challenges, particularly when there is uneven illumination on the objects. To address these challenges, a novel SOD algorithm is proposed, that employs manifold ranking on high-level features learned through a deep segmentation network. Unlike existing methods that use raw HSI as input, the proposed approach utilizes spectral gradient as input, mitigating the effects of spectral variation. The learned features are ranked using a graph-based manifold ranking approach, which enhances the robustness of the detection process. Experimental evaluation on the HS-SOD dataset demonstrates that the proposed approach performs better, achieving superior results compared to state-of-the-art methods, with a precision of 0.63, an f<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</inf>score of 0.51, and a Mean Absolute Error (MAE) of 0.08.