Title:  Uncertainty Estimation of Visual Attention Models using Spatiotemporal Analysis

Committee:

Dr. Ghassan AlRegib, ECE, Chair , Advisor

Dr. Biing Juang, ECE

Dr. David Anderson, ECE

Dr. Christopher Barnes, ECE

Dr. Berdinus Bras, ME

Abstract:

The objective of this research is two folds (i) analyze uncertainty in computational video saliency and (ii) design an effective uncertainty estimation algorithm tailored for video saliency detection. In computational video saliency detection, we highlight interesting regions or objects that might attract human attention when watching a video. Many video and image processing applications such as object segmentation, compression, and quality assessment utilize video saliency to efficiently reduce the dimensionality of the input videos and focus only on regions and objects that are interesting to human visual attention. However, there has been no explicit design of a saliency-based video processing framework nor analysis of the saliency maps reliability. In this dissertation research, we analyze eye tracking data and video content to discover general patterns of human visual attention that can be used for uncertainty estimation including map consistency, motion, and high-level saliency feature. Based on such analysis, we design a multi-factor uncertainty estimation algorithm and show its effectiveness in the application of video saliency detection.