Light field saliency vs. 2D saliency: A comparative study

In this paper, we present a comparative study of light field saliency and 2D saliency on light field dataset. Light field model exploits the focusness information obtained from light field at different depth levels using refocusing technique, which is suitable for handling challenging scenarios such as background clutter, similar foreground and background, and complex occlusions. Conventional saliency models only use 2D information (i.e., color, texture, etc.) on a single image and rely on assumptions about the properties of objects and backgrounds. Many of these techniques are fragile and may fail when the background is clutter or similar to the objects. We compare the accuracy of saliency detection between light field saliency and conventional 2D saliency on a challenging light field saliency dataset. The results on this dataset show that the effectiveness and reliability of light field information in saliency detection.