Experimental Analysis and Modeling of Digital Video Quality

University Of California-Santa Barbara, Santa Barbara CA

Investigators

Abstract

Abstract Experimental Analysis and Modeling of Digital Video Quality The cost of the production and transmission of digital video imagery is large and roughly proportional to the information in the images. Consequently, information is routinely deleted from video imagery to reduce cost. This deletion reduces the quality of the imagery as perceived by human viewers. The ultimate objective of this project is to quantify and minimize this loss of quality. As a step toward this goal, an understanding of how humans evaluate the quality of video images needs to be developed. To gain this understanding a series of experiments are being performed in which human observers describe and evaluate defects that they detect in video images of the kind that are typical in digital video applications. These images are processed so that they incorporate defects similar to those that occur in digital video. The results of these experiments are then used to develop an understanding of how humans process and evaluate video imagery. Image processing methods are being used to create sets of video defects that are quite similar to typical video defects, but are under much greater control. These defects occur at random times and places within normal video clips. They are treated as visual signals presented in a context of normal video imagery. Established methods of visual pattern detection and visual pattern appearance analysis are applied to the study of these defects. These include the following tasks: detecting the presence of a defect, rating the annoyance caused by a defect, and analyzing the perceived features of a defect. The results of these experiments will be used to develop an increasingly detailed model of video quality evaluation by human observers.