While the world we live in becomes more saturated with ubiquitous digital displays, and as the threshold for creating digital media continues to drop, image quality is an issue that concerns an increasingly large segment of the population. Higher resolutions, increased dynamic range, and faster frame rates put increasing demands on resources such as disk space and transmission bandwidth. Unfortunately, these resources are also needed for other functionalities of our digital devices and are often in short supply. To find new ways to optimize the production pipeline of visual media while maintaining a good image quality, more knowledge is required about how we perceive visual content. In this work, we examine how a specific viewing task or content affect the viewing behavior of an observer. We then examine how localized differences in image integrity affect the overall perceived quality. From these results we gain knowledge on how image quality should be optimized for a given viewing behavior. In addition, we show that for specific tasks there is a limit to the required content integrity. We investigate these research questions empirically using eye tracking to scan in real time how the viewing behavior changes under different tasks and for different content, while one of the tasks involved scoring image quality. Our results show that the viewing task and image content have a significant effect on the viewing behavior. We also find that the region of interest has a 5 times stronger effect on perceived quality in still images than the rest of the image. In videos, this effect is increased to 10 times. This finding can be utilized to optimize digital content once the region of interest is identified. We finally find that certain applications can mask degradations in image quality, making it redundant to allocate extra resources to maintaining content integrity.

Electrical Engineering, Mathematics and Computer Science

Intelligent Systems