VISUAL SALIENCY ANALYSIS AND APPLICATIONS

Abstract

Visual saliency refers to the preferential fixation on conspicuous or meaningful regions in a scene that have also been shown to correspond with important objects and their relationships. It is naturally built into the complex biological system to rapidly detect potential prey, predators, or mates in the real world. Visual saliency is also crucial for human visual experience and also relevant to many applications. Visual attention - particularly stimulus-driven, saliency-based attention - has been an active research field over the past decades. Many attention models are now available, which aside from lending theoretical contributions to other research areas, have given rise to successful applications in computer vision, mobile robotics, and cognitive systems. This thesis makes contributions on four aspects of visual saliency analysis, namely image re-attentionizing, 3d saliency, video saliency prediction, and action recognition. Each contribution has outperformed the state-of-the-art methods in different problems and validated the effectiveness of visual saliency. Beyond the aforementioned directions, we foresee more applications of visual saliency in image classification, image collage, video summarization and avatar thumbnailing.