使用CUDA加速的快速容積成像演算法

Abstract

容積成像演算法(Volume Rendering Algorithm)是一種將三維陣列資料視覺化的技術，廣泛應用在醫學和各種科學測量和模擬，根據不同的設備環境，有許多不同的實作方法，但由於龐大的陣列資料處理量，相當難以達成互動式繪圖的效果。本研究選擇其中一種不需依靠特定硬體，又具有高效率和高繪圖品質的演算法，剪裁-彎曲演算法來做三維繪圖演算法的改良，期望能達到及時繪圖的效果。 本研究所使用的繪圖函式庫VolPack是一套剪裁-彎曲演算法(Shear-Warp Algorithm)的實作軟體，基於剪裁彎曲演算法累加二維體積切片成就三維立體影像的特性，本論文改良體積切片累加成像的程序，利用Nvidia公司出的平行架構函式庫CUDA將演算法做平行化改良，目標是超越及時繪圖的門檻，也就是FPS=30(每秒30張)的繪圖速度。

Volume rendering is a technique for visualizing 3D array data. It has application in areas such as medical imaging and scientific visualization. According to the difference of the hardware environment, there are many kind of implementation, because of the massive array size, it is hard to achieve interactive visualization system. Shear-warp algorithm doesn’t depend on special hardware, it has high efficiency and high image quality. The research improve this algorithm, hope to accomplish an interactive visualization system. VolPack is a software implementation of shear-warp algorithm. Based on the special character of shear-warp algorithm that accumulate 2D image to 3D image, the research use CUDA library which is offer by Nvidia to improve the process for transcending the threshold of interactively rendering(FPS=30).