摘要
为实现射线图像复杂大背景下微小目标检测,研究强噪声、大灰度梯度下微小缺陷的分割方法。提出面向射线图像的视觉显著度模型,模拟人眼视觉注意机制,采用线扫描及自适应中央-周边差策略,以视觉显著度为尺度,通过特征图计算与融合、显著图获取等算法,从射线图像复杂背景中分割出注意区域;进一步通过显著度竞争标记排序各注意区域,并根据显著度阈值识别可疑缺陷区域,由此减少图像数据处理量,排除射线图像其他部分的干扰。提出以显著图上可疑区域的注意焦点为种子点,基于各点显著度的区域生长分割方法,实现了可疑区域图像中微小缺陷目标的准确提取。在复杂大背景X射线图像的实验中,准确提取出含有未知缺陷目标的区域,对微小目标的分割取得了较好效果,准确率达到96.1%,比传统方法高8%以上,证明了所提方法的有效性和适应性。
To achieve detection of tiny defect in radiographic images with complex background,the segmentation method of tiny defects was studied under the conditions of strong noise and large gray gradient background. The visual attention model for radiographic testing image was proposed,and its realization method was analyzed in detail. The human visual attention mechanism was simulated. The line scanning strategy and self-adapting central-peripheral difference strategy was adopted. Based on the vision saliency,the feature map and the saliency map were achieved,and visual attention region was segmented from radiographic images with complex background. Each visual attention region was marked and ordered with visual saliency competition. According to the saliency threshold, the suspicious region was identified. So the image data to be processed was reduced and the interference was discharged from other parts of radiographic testing image. Then attention focuses of the suspicious region was used as the seed point. Based on region growing and visual saliency,a segmentation method for tiny target was introduced to accurately extract tiny defects in suspicious region image. In the experiment about complex radiographic testing image with more tiny target objects,each area containing unknown defect was accurately extracted. Segmentation for tiny target achieved good results. The accuracy rate was 96. 1%,and it was 8% higher than that of the traditional method. The results prove the effectiveness and adaptability of the proposed method.
出处
《农业机械学报》
EI
CAS
CSCD
北大核心
2015年第7期365-371,共7页
Transactions of the Chinese Society for Agricultural Machinery
基金
重庆市基础与前沿研究计划基金资助项目(cstc2013jcyj A70009)
国家自然科学基金青年基金资助项目(51075419)
关键词
射线图像
微小缺陷
缺陷提取
视觉注意机制
视觉显著度
Radiographic image Tiny defect Defect detection Visual attention mechanism Visual saliency
