基于Hough变换的直线和圆检测的并行化研究-计算机软件与理论专业论文.docxVIP

摘要 人类对物体的识别主要是对物体外形的识别，图像形状检测在图像处理及模式识别中是很重 要的。Hough变换是种在图形图像处理中经常用到的形状检测方法，实现的是一种从图像空间到 参数空间的映射关系，具有一定的抗干扰能力、明了的几何解析性和易于实现并行处理等特点。 其最大的优点是检测几何形状的能力基本不受几何形状中间断点的干扰，而且不需要预先组合或 连接边缘点，可以广泛应用于孔径、纤维等自动检测中。 Hough变换存在着运算时间长的缺点，本文利用了并行处理这种解决海量数据计算的有效方 法来减少其运行时间。本文主要研究了：1)利用TBB(Threading Building Blocks)这种线程构建模块 在多核机上对Hough变换中可并行的部分进行并行化；2)基于CUDA(Compute Unified Device Architecture)这种并行计算平台，利用GPU(Graphic Processing Unit)的强大计算功能对Hough变换中 最耗时的部分进行计算。实验表明这两种方法对Hough变换的并行化都有很好的加速效果。 关键词：形状检测，Hough 变换，TBB，CUDA ，GPU Abstract The identification of objects’ profile is the most important way to recognize objects for human. It is very to detect shape features in the filed of image processing and pattern recognition. Hough transform working as a common method for shape detection in image processing constructs a mapping from image space to parameter space, which is characterized by geometric analyticity, immunity from interference in a sense and great facility for parallelization, etc. Its principal advantages can be listed as follows: the capacity for geometric shape detection which almost can not be disturbed by intermediate breakpoints in geometric shape, no need to assemble or link edge points beforehand and widespread application in the automatic detection for fibre and aperture. However, the severe short coming of too long computation time hold back Hough transfrom. So parallel processing, a particllarly effective method, is introduced to cope with the mass data calculation in order to speed up the computation. In this paper, two different means of parallelization is given: One method is to make use of TBB, to parallelize the parallelizable part in Hough transform on multicore machine. The other one is to utilize the powerful computing function of GPU based on CUDA, to parallelize the most time-consuming part. Experimental result shows that these two different methods both works very well on the accelerating effection for the parallelization of Hough transform. Keyword: shape d