Rendering egomotion using dynamic textures

This thesis is focused on addressing some of the limitations present in the spectral synthesis technique introduced by Zhang to render dynamic textures resembling falling snow. We improved the speed and the memory requirements of the rendering process by replacing the 3D inverse Fourier transform used in the rendering technique by a sequence of 2D inverse Fourier transforms. One of the issues with the rendering technique was that it didn't take into account the variations across the motion field that are visible when the observer undergoes non-lateral motion. In order to incorporate the motion field effects caused by egomotion (i.e. the motion of the observer) we devised a tiling technique which involved dividing the view field into square image tiles, rendering each tile with a unique direction of motion and speed range (depending on the underlying motion field) and then blending the tiles together. The tile boundaries are not visible in this case because of certain constraints present in the human visual system (HVS) with regard to processing such motion gradients or discontinuities. Finally, based on the constraints present in the HVS in processing multiple moving layers, we show that the original model which renders a large number of overlapping transparent layers is an overkill and demonstrate a simplified model which renders just a limited number of layers.