(A) study on texture mapping algorithms for rendering hardware

Abstract

There have been a lot of progresses in rendering hardware and at present, the quality of texture mapped image by commodity rendering hardware looks very similar to real objects. It is mainly due to texture mapping. Therefore, the texture mapping in rendering hardware becomes very important for the high performance. In this dissertation, three texture mapping algorithms for rendering hardware are proposed. The first texture mapping algorithm is to reduce the bandwidth of texture memory. To reduce the I/O bandwidth of the texture memory, a novel method using a map, called hint map, is proposed. The hint-map is devised based on the fact that the number of texels within a block can be reduced by the variance of texels in the block. A hint bit in the hint map for a block specifies that the variance of texels within a block is less than a threshold value. During fetching a block from texture memory, if a hint bit for a block is 1, all texels of the block are fetched. Otherwise, one representative texel is fetched. Therefore, if a texture image has many blocks where the variance of texels is small, the I/O bandwidth consumed in texture mapping by using the texture image can be reduced. This method is evaluated using several images. The reduction of the bandwidth using the hint-map ranges from 85% to 97%. The second proposed algorithm is to enhance the quality of the MIPmapped image. It is the extension of the original footprint assembly algorithm\cite{Strasser96} and can produce clearer texture mapped image than footprint assembly. In spite of the fact that the shape of footprint is an ellipse where one end is inflated and the other end is deflated, the footprint assembly approximates the shape of footprint by considering only the major and minor axis of ellipse. In other words, it does not consider the variation of level of detail values along the major axis. Therefore, there are the quality degradation of texture mapped image by the footprint assembly. How...