Research On Application Of Spatial Information In Transfer Function Design

With the development of information technology, the amount of3D volume dataobtained from either scientific computing or sensing device collection has grownsignificantly, which has resulted in an increased need to interpret/explore these data.Direct volume rendering(DVR) is an important method for volumeexploration/visualization. Transfer function is one of the most commonly used tools involume rendering, which determines the final DVR image by mapping the data intocolor and opacity specifications. However, it’s still a challenge to find some propertransfer function in general, for the complexity of various volume data and thedifficulty of feature extraction.In traditional transfer function design, the scalar and gradient magnitude of thevolume data would be considered usually, while spatial information sometimes beoverlooked. On the other hand, manual transfer function design is too critical for theusers to interact. To solve these problems separately, this dissertation proposes twomethods applying spatial information into transfer function design.(1) Gaussian transfer function design based on slice painting. The procedure is asfollows. It obtains user’s paintings on the slice images and maps voxels on thesepaintings to corresponding points in high demensional affine space. Then aGaussian radiual basis function interpolation is established and solved bynon-negative least squares algorithm. Finally it uses this interpolation to classifythe voxels. This method puts spatial information into the parameter space oftransfer function, then enhances its capacity of classification.(2) Semi-automatic transfer function generation by local clustering of histogram. Itaims for a facilitating interaction by semi-automatic transfer function generation.The provided method uses the result of local clustering of the volume’sscalar-gradient histogram, which involes the mean and variance value of voxels’spatial coordinates that belongs to some scatter point of the histogram. Itdemands for fewer manipulations to gain an understanding rendering image withthis method.These methods have been implemented with GPU-accelerated ray castingalgorithm using Phong illumination model. Experimental results of several volumedata prove the methods’ effectivities.