Increasing ubiquity of computer graphics demands high levels of realism. If computer images are to be useful beyond entertainment and simulation, then realism is the key. An important, yet largely unanswered question, is how do we measure the realism of computer generated imagery?

One approach, the approach taken in my research, is to use psychophysics to study the relationship between stimulus (image) and response (visual perception). By conducting a series of psychophysical experiments it is possible to assess the fidelity of graphical reconstruction of real scenes. Experimentation also shows that given a real scene and a faithful synthetic representation of that scene, the visual response function in both cases is similar. This gives a measure of the realism of the synthetic image. These experiments were conducted using a specific set of images – how can these results be generalised to other images, or if they cant how should I design future experiments which will yield results which can be applied to other image scenarios.

Setting up and running experiments is a (painfully) slow process, so it is important to get it right first time or you end up with a set of useless results. I am interested to learn how others go about conducting perceptual experiments to ensure robust results, including designing the experiment, choosing/setting the task, participants, analyses and generalisation of results.

Also, the role of eye movements in image quality assessments is largely unexplored. Can the pattern of eye-movements be used as an indication of the realism in rendered images? To investigate this, eye-movements were monitored while participants compared images of varying quality, including a photograph. This aims to establish if there is a relationship between the pattern of eye-movements and fixations with the quality of a rendered image.

This project is in its infancy, however, preliminary results show that there are indeed common fixation areas among images of similar quality. The figure below shows the patterns recorded for one participant. Images are shown in the rating order given by that participant (on a scale of 0-10). It can be seen that gaze is concentrated primarily in the centre of the image, with fixations mostly on the objects in direct lighting.