An assessment model and implementation of stereo image quality

Abstract

In the past decade, many display hardware manufacturers have initiated research into the construction of stereo display devices. Currently, the use of such displays is limited to the computer-aided design; research, military and medical applications. However, it is anticipated that as display hardware becomes cheaper, gaming companies and desktop application software developers will realise the potential of using stereo to provide more realistic user experiences. To provide realistic stereo user experience it is necessary to utilise good quality stereo images in addition to suitable hardware. The growth of the Internet has resulted in an increase in the availability of stereo images. However, most have been captured using uncontrolled procedures and have questionable quality. The quality of stereo images is important since the viewing of poor quality stereo images can result in adverse viewing effects. A formal definition of stereo quality has not been achieved in current day research. This means that the factors which cause a stereo image to be perceived as poor quality have not been defined nor is a system available to detect its occurrence. This thesis attempts to address this problem by postulating a definition of stereo image quality based on detecting level of excess disparity levels, intensity differences and the occurrence of frame cancellation. An implementation system able to detect these identified factors is discussed and formulated. The developed system is utilised to test 14 stereo images of varying quality levels. The results of these tests are reported and are used to evaluated and refine the system. Using this image analysis, benchmarks for natural intensity difference in images, changes due to JPEG compression and comparisons with generated and ground truth disparity maps are formulated. Additionally, an