Abstract: In order to detect bubble defects in Digital Radiography(DR) images of railway castings automatically, snake model is applied to segment these defects, and the method of getting the initial contour of Snake model is improved. The barycenter of every bubble is achieved by using automatic thresholding and region growing method. A method based on radial is applied to get the series of initial control-points of Snake model. These control-points are converged by Greedy algorithm. Experimental results from simulated images detecting show that the minimal size which this method can detect is 3×3 pixels, and the initial contour may be able to converge to the concave areas. Experimental results from real DR images detecting prove that this algorithm can detect the contour of bubbles in detecting region exactly, and there will be no false defects in the results, and also this algorithm improves the automation of defects detecting.

Key words: Snake model, bubble defects, Digital Radiography(DR) images, railway castings

摘要： 研究Snake模型在铁路铸件数字化辐射成像(DR)图像气孔缺陷自动检测中的应用，改进初始轮廓点的获取方法。综合运用阈值分割和区域生长方法得到各气孔的重心，依次对其采用一种射线法得到初始控制点，对初始控制点进行收敛和拟合。对仿真图像的实验结果表明，最小检测尺寸为3×3，能较好地收敛到目标的凹陷区域；对实际铁路铸件DR图像检测的实验结果表明，该方法能准确得到检测区域内多个气孔缺陷的轮廓，不会检测出伪缺陷，具有较高的自动化程度。

关键词: Snake模型, 气孔缺陷, 数字化辐射成像图像, 铁路铸件

CLC Number: 

• TP391