TIGGER: A Texture-Illumination Guided Global Energy Response Model for Illumination Robust Object Saliency
Sara Greenberg, Audrey G. Chung, Brendan Chwyl, Alexander M. Wong · 2016
Global saliency is an important aspect of many computer and robotic vision tasks, and with the increased interest infields such as autonomous navigation, a significant area of research. A challenging aspect of modelling global saliency in practical applications is the presence of varying or non-uniform illumination conditions. Many current models fail to accurately detect salient regions in non-uniform illumination conditions and often produce different saliency maps for the same image under changing illumination. In this paper, we propose a novel model for illumination robust global saliency. For a given input image, texture-illumination guided energy responses (TIGERs) are computed at different scales using a novel multi-scale extension of TIGER. To acquire these responses, image intensity is modelled as the summation of the low frequency illumination component and the high frequency texture component. A captured image is disassociated into these components via Bayesian minimization, with the required posterior probability estimated through an importance-weighted Monte Carlo sampling approach. The texture-illumination guided global energy response (TIGGER) is computed as the aggregate sum of TIGERs across all scales. The global saliency map is obtained via a k-means clustering-based region adjacency graph (RAG) model. Experimental results produce global saliency maps with improved performance in non-uniform lighting conditions and greater consistency when compared to other state-of-the-art methods.