Computer Generation of Integral Images using Interpolative Shading Techniques

Research to produce artificial 3D images that duplicates the human stereovision has been ongoing for hundreds of years. What has taken millions of years to evolve in humans is proving elusive even for present day technological advancements. The difficulties are compounded when real-time generation is contemplated. The problem is one of depth. When perceiving the world around us it has been shown that the sense of depth is the result of many different factors. These can be described as monocular and binocular. Monocular depth cues include overlapping or occlusion, shading and shadows, texture etc. Another monocular cue is accommodation (and binocular to some extent) where the focal length of the crystalline lens is adjusted to view an image. The important binocular cues are convergence and parallax. Convergence allows the observer to judge distance by the difference in angle between the viewing axes of left and right eyes when both are focussing on a point. Parallax relates to the fact that each eye sees a slightly shifted view of the image. If a system can be produced that requires the observer to use all of these cues, as when viewing the real world, then the transition to and from viewing a 3D display will be seamless. However, for many 3D imaging techniques, which current work is primarily directed towards, this is not the case and raises a serious issue of viewer comfort. Researchers worldwide, in university and industry, are pursuing their approaches in the development of 3D systems, and physiological disturbances that can cause nausea in some observers will not be acceptable. The ideal 3D system would require, as minimum, accurate depth reproduction, multiviewer capability, and all-round seamless viewing. The necessity not to wear stereoscopic or polarising glasses would be ideal and lack of viewer fatigue essential. Finally, for whatever the use of the system, be it CAD, medical, scientific visualisation, remote inspection etc on the one hand, or consumer markets such as 3D video games and 3DTV on the other, the system has to be relatively inexpensive. Integral photography is a ‘real camera’ system that attempts to comply with this ideal; it was invented in 1908 but due to technological reasons was not capable of being a useful autostereoscopic system. However, more recently, along with advances in technology, it is becoming a more attractive proposition for those interested in developing a suitable system for 3DTV. The fast computer generation of integral images is the subject of this thesis; the adjective ‘fast’ being used to distinguish it from the much slower technique of ray tracing integral images. These two techniques are the standard in monoscopic computer graphics whereby ray tracing generates photo-realistic images and the fast forward geometric approach that uses interpolative shading techniques is the method used for real-time generation. Before this present work began it was not known if it was possible to create volumetric integral images using a similar fast approach as that employed by standard computer graphics, but it soon became apparent that it would be successful and hence a valuable contribution in this area. Presented herein is a full description of the development of two derived methods for producing rendered integral image animations using interpolative shading. The main body of the work is the development of code to put these methods into practice along with many observations and discoveries that the author came across during this task.