The objective of this work is to develop mathematical models (using machine learning regression) that can mimic human ability to perceive... Show moreThe objective of this work is to develop mathematical models (using machine learning regression) that can mimic human ability to perceive motion of a small object in an image sequence or a video. The motivation for this work comes from the diagnostic cardiac imaging where a small deformation of a cardiac wall motion represents a signi cant diagnostic marker. First a brief overview of the state of the art in image and video quality assessment is given. This overview also points out a need for a new task based quality metrics which can better quantify, subjective, image sequence quality under various degradations, like blur and noise. Next a study is designed to measure human observers motion perception under various degradation models. Later, the results from this study are analyzed to detect which image sequence features are the most relevant for motion perception and the development of a mathematical model aiming to emulate humans. The chosen features are based on so called visual attention and estimated object motion. Since the computation requirement to calculate visual attention and estimated object motion are considerable we also present a fast parallel implementations based on graphical processing units using NVIDIA Compute Uni ed Device Architecture. The preliminary results indicate that proposed machine learning regression models with the use of the visual attention and estimated object motion can accurately predict human motion perception. M.S. in Electrical Engineering, May 2011 Show less
