Date of Graduation

2005

Document Type

Thesis

Degree Type

MS

Committee Chair

Arun Ross

Abstract

The ballistic movement of the human eye to comprehend a scene is known as saccades. Saccadic eye movements have been studied by researchers to understand their role in visual cognition. These eye movements are key to providing the brain with visual information that aids in the interpretation and perception of the scene being viewed. Though saccades are primarily task driven, it has been theorized that the first few saccades are triggered by low-level image properties that are inherent in a scene. Determining and evaluating these properties would enhance the understanding of visual cognition and, thus, benefit the development of “eye-friendly” visual applications. In this report, we examine image information both at the pixel level and the object level in order to determine their relevance to saccadic activity. Pixel level analysis was carried out under the premise that, in the lack of visual cognition, saccadic activity is triggered by low-level image properties such as color, texture, edges, etc. Object level analysis was conducted under the assumption that some amount of visual cognition precedes saccadic activity. Results, based on a supervised learning algorithm, indicate that the hue and saturation features of an image play a significant role in directing the first few saccades of the eye when it encounters a novel scene. Furthermore, preliminary observations indicate the possibility of utilizing depth perception for predicting fixation targets.

Download

Share

COinS