【摘要】 在工业x光图像安全检测中,缺陷在某种尺度下可以看作是屋脊边缘。从多尺度边缘检测的角度分析得出:在一定尺度下,圆形缺陷在两个方向上都看作是屋脊边缘,而条形缺陷只在一个方向上被看作是屋脊边缘。利用小波多尺度边缘检测的两个分量来确定两个方向是否都存在屋脊边缘,从而区分开这两种形状缺陷。使用LOG算子提取圆形缺陷,使用方向可调滤波器提取条形缺陷,根据检测到的缺陷的形状参数来对缺陷进行分类判级。实验表明,该算法有很好的检测效果。更多还原

【Abstract】 In safety detection of industrial X-Ray image, the defect is considered as crest edge in certain scale. From multiscale edges detection,the conclusion is drawn that in certain scale, circular defect is considered as crest edge in two directions, and bar defect in one direction. Two parameters from wavelet multiscale are used to confirm whether crest edge in two directions exist or not, then the circular defects and bar defects can be distinguished. Different operators are used to inspect the two types defect, LOG operator for the circular defect, and steerable filter operator for bar defects, then it can classify the defects based on detected shape parameters of defects. Experiments show this algorithm is efficient.更多还原