Understanding Visual Saliency and Perception of Digital Media
Loading...
Date
2015
Authors
Rahman, Shafin
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Eye movements are directed by both high-level considerations such as contextual guidance or experience and also by observed visual patterns. A bright light or high contrast image structure will tend to draw an observer's attention, independent of task. The extent to which a visual pattern attracts attention in this manner is often described according to a quantitative measure of visual saliency. Although many computational models of visual saliency have been published in last two decades, current literature is mostly limited to the maximization of performance in predicting saliency. Apart from this common challenge, we address some new aspects of image understanding relating to saliency. For example, we explore the relationship between natural image statistics and human visual attention, measure the extent of center bias produced by saliency algorithms and develop ways to remove such bias, relate different perceptual task like free viewing, object search, saliency search and explicit judgment, and predict different image level subjective ratings. In this thesis, we present solutions to aforementioned problems for a better basic understanding of visual saliency while also producing some novel avenues of analysis that inform on perception of digital media.
Description
Keywords
Visual Saliency