在表面缺陷自动光学检测图像处理中,需要对每个缺陷进行标记,便于后续判别缺陷的类型、大小、位置、方向等Blob特性.研究了应用Hotelling变换(霍特林变换)求出缺陷的主轴方向和最小外接矩形,沿着主轴方向将缺陷标记出来的算法,并给出了应用NVIDIA的CUDA架构对标记算法进行加速实现的方法.研究结果表明,根据图像缺陷大小的不同,可以取得5～10倍的加速实现存记.%Automatic optical detection of surface defects in image processing, in order to judge the defect's Blob features, such as type, size, location and direction, need to mark the every defect Used the Ho telling transform to find the principal axis and minimum enclosing rectangle of the defect Along the principal axis, the defect is marked in minimum enclosing rectangle. The transform algorithm based on NVIDIA's CUDA architecture is accelerated. According the size of the defects, the algorithm is accelerated to (5 ~ 10) times the speed.
展开▼
