摘要
根据锥束CT序列图像各向同性及缺陷实体在序列图像层间位移和变化较小的特点,提出了一种适合锥束CT序列图像的三维缺陷检测方法。首先结合多目标跟踪思想,利用三维连通区域标记算法提取三维缺陷实体,并建立缺陷对应关系哈希表,以解决缺陷检测中目标轨迹的分叉;然后根据噪声目标的固有特性,对虚假缺陷信息进行有效删除,最终可获得真实缺陷目标实体,缺陷提取精度可达到像素。通过对空心涡轮叶片蜡模锥束CT图像进行实验,结果表明。该方法能准确地提取有较强噪声影响的序列图像的三维缺陷。
According to the isotropy of serial slice images of Cone-Beam Computed Tomography （CBCT） and the small displacement and little shape change of defect solids between slice images, a method of 3D defects detection for CBCT serial slice images is proposed. Firstly, the defect solids are labeled by multi-object tracking and 3D connective region extraction. The Hash table of correlative defects is establishes to solve the fork trajectory of defect targets. Then the fake defects are deleted by the inherent particularities of noise targets, so the final real defect targets are obtained and classified which precision of extracted defects is 3 × 3 × 3 pixels. This method is adopted to processing the CBCT serial slice images of wax model of hollow turbine blade. The experiment result shows that the 3D defects in the serial images corrupted by noises are extracted exactly.
出处
《光学技术》
EI
CAS
CSCD
北大核心
2008年第6期914-917,923,共5页
Optical Technique
基金
国家科技支撑计划重点资助项目(2006BAF04B02)
西北工业大学青年科技创新基金资助项目(W016221)
关键词
锥束CT
序列图像
三维缺陷
多目标跟踪
cone-beam computed tomography
serial slice images
3D defect
multi-object tracking
