Analysis, control and evaluation of image generation in volume rendering.

摘要

The last few years has spawned much interest in volume visualization. The methods of volume visualization have been used to analyze and render on a computer display 3D datasets obtained from a variety of sources including medical scanners and results of simulation of physical and synthetic phenomenon. Much work has been reported in the development of basic rendering algorithms and optimizing them. However, there has been little effort at developing highly accurate algorithms to analyze and render volume datasets. Also, there is no effort to measure the usefulness of more accurate schemes through an examination of the final rendered 2D image. In this thesis we examine the basic volume rendering pipeline and identify different stages which can gain from more accurate treatment and thus result in a more realistic renditions. Among the different stages of the pipeline we particularly examine the classification and classification stages. Also, we propose image comparison metrics which can be used to guide a typical image generation effort.; Towards the development of better classification schemes, we also develop a new method of sub-band combining to identify structures in an image. This method is based on the orthogonal wavelet transform. The ability of compactly supported wavelets to detect singularities (edges, ridges, corners) more ably than traditional methods is used to locate structures in an image. The result of identification is not region-wise but pixel-wise allowing better local control of structures. Our assignment of saliency values are novel in that they measure the presence of frequencies at all scales. Thus singularities like a step are rightly assigned very high saliency values.; To characterize reconstruction operations better, we propose metrics to measure the error of reconstruction in the spatial domain. Also, the errors proposed are local in nature. The latter aspect is exploited to develop position and data dependent adaptive filtering schemes.; Finally, to be able to measure the effectiveness of the reconstruction and better classification (and other improvements) we develop image comparison metrics. These metrics are again based on the wavelet transform. One of the metrics we describe is directly derived from the combining algorithm and measures the structure content in an image. The second metric goes beyond the first metric in that it measures the structure content and also includes the response of the Human Visual System.