中文摘要：

针对CT体数据的多尺度特征点检测计算量大、耗时长的问题,提出一种三维中心环绕特征快速检测算法。设计三维中心环绕特征检测子,结合三维积分图像快速生成图像的尺度空间,同时利用三维Harris边缘判定准则去除边缘点,增强特征点的稳定性。实验结果表明,相比于经典的三维Do G和SURF检测子,算法计算时间显著降低（检测时间约为三维Do G检测子的1/8,三维SURF检测子的1/2）,同时相比于三维SURF检测子,特征点检测重复率也有一定程度的提高。最后,对三维中心环绕特征检测算法进行并行性分析,并分别从尺度空间生成和特征点搜索及边缘抑制两部分进行CUDA并行加速。实验结果表明,经CUDA加速后,算法能得到10倍左右的加速比,特征点检测过程耗时基本达到实际应用需求。

英文摘要：

Aiming at the problem of large-computing and time-consuming of multi-scale feature detection for 3D volumetric data, this paper proposed a fast center surround feature detection algorithm. This algorithm firstly designed a 3 D center surround feature detector. And with the help of 3D integral image, it would build the scale space of 3D volumetric data more quickly. Meantime,it also used 3D Harris edge criterion to eliminate edge points to enhance the stability of feature points. Experiments show that the time this algorithm taken obviously decreases compared with the classical 3D DoG and SURF detector and the repetition rate of this algorithm increases compared with 3D SURF detector to some extent under the same conditions. At last, this paper analyzed the parallelization of the algorithm and used CUDA to accelerate the detector from scale space generation and feature point searching as well as edge suppression aspects. Experiments show that the speedup ratio of this algorithm can reach about 10 with CUDA and the feature detector can satisfy the practical application after acceleration eventually.