Dipmap: A new texture filtering method for texture mapping

High image quality and efficiency are usually conflicting goals in texture mapping, especially for real-time rendering. Mipmap is a typical texture filtering method for texture mapping. Due to simplicity and efficiency, it has been popularly used in current graphics hardware. However, its implementation efficiency is at the cost of image rendering quality. In this paper, we first present an improved sampling scheme using the same Mipmap pyramid. In this scheme, for each screen pixel in the scanline, instead of just using the texture scaling ratio in a single direction to determine the Mipmap pyramid level for sampling, we take a row sampling and a column sampling respectively in different Mipmap pyramid levels to generate its value. These two pyramid levels are determined independently by the texture scaling ratios in the row and column directions. The double sampling scheme can overcome the aliasing and overblurring problems caused by just sampling using the texture scaling ratio in a single direction. Motivated by this, we propose a new texture filtering method called Dipmap (Double Mipmap) to further improve image rendering quality. In Dipmap, we establish a row pyramid for row sampling and a column pyramid for column sampling to make these samplings completely independent of each other. Experiments show that Dipmap can deliver much improved image rendering quality than the state-of-the-art texture filtering methods like Mipmap, Ripmap and FAST, while being nearly as efficient as Mipmap.