Fast volume rendering using adaptive block subdivision

Abstract

The rapidly increasing prevalence of computer to interpret huge amount of sensory and numerical deatasets has created demands for a variety of visualization and perception techniques. One of them is the Volume Bisualization to give a 2D presentation of the scanned object or some other data features of 3D scalar datasets. Another one is the Segmentation and Classification that extracts regions of interest from given input datasets. To achieve the fully automatic interpretation of datasets by computer, visualization method of 3D volumetric scalar datasets and an automatic segmentation method of 2D inages are studied in this thesis. After choosing ray-casting algorithm as the basic visualization algorithm due to its large versatility, we propose a fast volume rendering scheme that greatly reduces the overhead of ray-block intersection calculation by uniform block merging approach. It is shown that our adaptive approach is effective in optimizing volume coherency by taking advantage of two approaches: Hierarchical space partitioning and uniform space partitionong. We first apply a uniform space subdivision for constructing the adaptively subdivided volume. Then coherent uniform blocks are merged to generate adaptive-sized blocks that are efficient for leaping space. To accomplish this, an efficient block merging algorithm based on a boundary following and scan-filling strategy is devised. Experimental results show that our fast volume rendering scheme is superior to precedent schemes in constructing a data structure. The construction speed is more than three times faster than a hierarchical data structure, the octree. The performance gain in rendering timing over the octree is 25% to 70%. Additional advantages, such as suitability for animation and large-scale volume rendering, are alos idenfified. In addition to the volume visualization in the part I of the thesis, automatic object segmentation and recognition are studied in the part II. An effective automatic fac...