摘要
通过改进GPU的渲染到纹理策略加速锥束反投影计算,提高了圆弧轨迹Katsevich锥束CT重建算法的执行效率.在CPU中完成投影数据的滤波处理,并将其重排为四通道浮点纹理载入GPU内存;绑定2个浮点输出纹理到帧缓冲对象的不同绑定点,用于将渲染结果输出到纹理,并同时完成反投影数据的累加计算;将矩形网格编码为四叉树结构用于加速物体切片的绘制.实验结果表明,文中算法在较大锥角时仍能重建出高质量图像,并获得接近于实时的重建速度.
The paper accelerates cone-beam back projection computation by improving GPU technique of render-to-texture, and enhances performance of Katsevich-type CT reconstruction algorithm for single circular arc trajectory. The projection data are filtered on CPU, and then rearranged and downloaded to the graphics hardware as the RGBA four-channel textures. Two floating output textures are bound to different attachment points of a frame buffer object, which enables the rendering results to be written to the textures and added with the previous back projection data. Rectangle grids are coded as quadtree structure applied to speed up rendering of object slices. Experimental results show that under the condition of big cone angle the proposed method can obtain high quality reconstructed images in near real-time reconstruction speed.
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
2008年第7期919-924,共6页
Journal of Computer-Aided Design & Computer Graphics
基金
辽宁省高等学校科学研究青年基金(2004F057)
关键词
图形处理器
浮点纹理
Katsevich锥束重建
圆弧轨迹
graphics processing unit
floating texture
Katsevich cone-beam reconstruction
circular arc trajectory
