Improving Efficiency for CUDA-based Volume Rendering by Combining Segmentation and Modified Sampling Strategies

The objective of this paper is to present a speed-up method to improve the rendering speed of ray casting at the same time obtaining high-quality images. Ray casting is the most commonly used volume rendering algorithm, and suitable for parallel processing. In order to improve the efficiency of parallel processing, the latest platform-Compute Unified Device Architecture (CUDA) is used. The speed-up method uses improved workload allocation and sampling strategies according to CUDA features. To implement this method, the optimal number of segments of each ray is dynamically selected based on the change of the corresponding visual angle, and each segment is processed by a distinct thread processor. In addition, for each segment, we apply different sampling quantity and density according to the distance weight. Rendering speed results show that our method achieves an average 70% improvement in terms of speed, and even 145% increase in some special cases, compared to conventional ray casting on Graphics Processing Unit (GPU). Speed-up ratio shows that this method can effectively improve the factors that influence efficiency of rendering. Excellent rendering performance makes this method contribute to real-time 3-D reconstruction