CUDA加速工业DR图像分割

摘要

对DR(digtal radiography,数字辐射成像)图像进行分割是工业DR图像处理中一项重要内容.C-V算法对DR图像分割效果较好,但该算法计算量大,在工业应用中达不到实时处理要求.结合高性价比CUDA技术实现C-V算法对DR图像分割并行化,并采用共享内存技术、独立计算与合并计算结合的方法,较大地提高了C-V方法的计算效率.对实际工业DR图像分割实验结果显示,该方法加速比可达到32～44倍,表明使用CUDA并行化C-V方法分割DR图像高效可行.%The segmentation of industrial DR image is very important to industrial image processing. C-V method has some advanage to industrial DR image segmentation. Due to higher complexity, C-V method is rarely applied in a timecritical industrial environment. This paper presented an efficient parallel implementation of DR image segmentation by C-V method with CUDA, and used a method that separated data computing merged in corresponding thread by shared memory. Finolly, compared the results to the, current applications on image processing workstation, the implementation exhibited a speedup of 32 to 44 times compared to a state-of-the-art CPU.