Compression of computer animation frames

This thesis presents compression algorithms for computer generated animation. Computer animation is defined as a sequence of still images which are rendered based on the information in an animation script. From computer graphics theory there is a geometric transform matrix for each object which relates any point in the object model space to image space. For an animation sequence, it is possible to compute a motion prediction matrix for each object that transforms each pixel from one frame into another frame. We developed motion compensated compression algorithms based on the motion prediction matrix.; The compression system is composed of a geometric data generator and a color data coder. The geometric data, i.e., object number and depth, at each pixel are provided to the color data coder for the motion prediction of the color data. There are two approaches with regard to the geometric data generation: a rendering and a compression method. We develop a motion predicted compression algorithm for the geometric data. The motion prediction of geometric data is achieved by transforming each planar pixel square. Also we propose a direction coding scheme to compress the geometric data of unmatched pixels using the neighboring pixel data. Test results demonstrate that the compression algorithm encodes the geometric data at about one to two bits/pixel.; With regard to color data compression, we develop a lossless and a lossy compression algorithm. Both compression methods are based on a motion prediction transform. In the lossless method, we propose two modified motion prediction methods to improve the compression gain. First, motion prediction with DPCM or MDPCM reduces the low frequency residue errors which generally occur due to the changes of shading or shadows between frames. Second, ratio prediction improves the prediction in high spatial frequency texture patterns. We also propose a block predictor switching algorithm (BPS), which uses an optimal predictor set for each block of pixels. Experiments show the new algorithm outperforms existing methods roughly by a factor of two in compression gain.; We develop a lossy compression algorithm which combines the new motion prediction method and the DCT coding used for the MPEG algorithm. Furthermore, we propose an edge weighted DCT coding scheme based on the error masking effect by object edges. This method significantly reduces the bit rate for geometrically complex images without any loss of perceptual quality. Experiments demonstrate that new method reduces the bit rate by 40 to 50% for the same perceptual quality or improves the perceptual quality by more than three dB in SNR for the same bit rate.