Key Techniques Study On Real-time Volume Rendering

Volume visualization especially volume rendering has been widely used in many areas of society such as medicine, meteorology, geology and so on. It is the most important and rapid developed techniques in the visualization of three-dimensional data sets in recent years. Several key techniques in volume rendering including accelerated strategies and classification design are mainly studied in this paper.Ray casting is most popular method of rendering high quality images from volume data. Its main shortcoming has been the high computational expense and low rendering rate. Meanwhile GPU (Graphics Processing Unit) is a powerful accelerated tool for its massive floating-point computational capabilities and highly parallel computing architecture. Therefore, a GPU-base ray casting algorithm is presented to take advantage of benefits from modern programmable GPU. The main idea of the algorithm is to reduce data transformation between CPU and GPU by finishing all resampling process along one ray in single rendering stage. Furthermore, traditional accelerated techniques of early ray terminal and empty space skipping are combined into the algorithm on GPU platform. It is proved that GPU-based ray casting could produce high quality results with real-time rendering rate for the volume data size within available video memory.According to the analysis of consistency in volume data, a segment-based ray casting is proposed in this paper. The new method achieves accelerating goal by combining sequential resamplers with similar values into one segment and substituting resampling segments for resampling points to be the composition units. Therefore the overall number of resampling and composition process is much reduced. The algorithm is verified efficient by both theoretical deduction of rendering equation and practical implementation on CPU and GPU. The CPU version is implemented to determine the recommended values of parameters and GPU version is used for further optimization.Transfer function plays an important role in volume rendering. As an improvement, a transfer function adjustment method based on extended sampler model is demonstrated in this paper. The model expands each sampler to a segment which contains several sample points. The original sampler is one of the sampler points and often locates in center of the segment. The value relationships between known sample points and unknown sample points are defined by certain template. By adopting the model with various templates, the transfer function is flexibly revised to perform better results with clearer boundaries.Besides optimizing strategy, another mapping rule for transfer function based on acting forces model is also introduced. In the acting force model, the resamplers are not the points interpolated at even interval along the ray but the first point in each coherent segment which is made up of sequential original resamplers with similar values. Therefore, the distances between neighboring resamplers are different according to local data coherence. Each resampler is treated as a particle which would move along the direction of acting force from the neighbor with an initial speed in the same direction. By applying the model to transfer function, the multi-dimension transfer function is degraded to one-dimension one by unifying several data features such as resampler's value, gradient and resampling interval to one physical concept kinetic energy. It is proved that the proposed method could produce flexible and good results with easy adjustment operations.