Study On Weak Edge Detection Of Industrial CT Image

CT technique is considered as an advanced detecting mean in nondestructivetesting field, and has been widely used in aviation, petroleum, machinery, automotive,mining and other fields. In the actual engineering, because of properties of interiordefect causes no clearly edges, such as air holes and loosens of casting, or surface andscanning fault plane are not vertical when some detected parts proceed with CTscanning as well as the volume effect of CT scan image, which causes the gray of CTimage edge has wider transition zone to form so-called gradient edge or weak edge.Weak edge detection is one of difficult problem for CT image segmentation anddetection. This paper focuses on the weak edge detection of CT image, which analysesthe advantages and disadvantages of mainstream image segmentation method as well astheir applicable scope, emphatically researches the watershed image segmentationmethod and applies it into the detection for weak edge of CT image. This is completedas follows:First, as for “over-segmentation” problem of watershed, the paper proposes theuse of morphological opening-closed reconstruction operations to filter original image,then to remove the high gray-scale and low-gray level details that are smaller thanstructural elements, and to ensure that the edge information of the original image cannotoffset or lose. The experiment shows that the filtered image reuses watershed-basedsegmentation method; it will not appear over-segmentation phenomenon.Second, it analyzes the causes of volume effect and its impact on image edge, dueto two unique volume effects of CT system cause typical workpiece’s cone and spherewith CT image weak edge, it conducts actual measurement and analysis, and tests theLeast Squares Approximation after adopting morphological open-closed reconstructionfiltering and watershed segmentation method, which can make the edge detectionaccuracy to the sub-pixel level. Meantime, in accordance with the segmentation resultson the volume effect and slice thickness, the relationship between the curve slope of theworkpiece axial edge and testing errors, it conducts a qualitative analysis.Third, after analysis of a common defect detection methods and industrial CTimage noise, according to the characteristics of noise and defects in CT image, it usesmorphological opening-closing reconstruction operator to remove image noise, whichcan better keep the basic shape characteristics and edge information of the defects in the original image, and better ensure the processed image edge will not offset; then reusethe method of image segmentation based on watershed, and finally fill the defect area,the defect area will be fully extracted, and then comparing it with four methods that areused currently to show the operation adopted in this paper has a better advantage in theextraction of weak contour in defected area.In conclusion, this paper conducts systematic research according to watershed-based image segmentation method, and the method is successfully applied to the weakedge detection in industrial CT images, it verifies accuracy and use range of edgedetection through experiment. Compared with the other methods, when this method isapplied to defect detection of industrial CT images, it can extract a connected andclosed defect edge, and achieve fast and accurate segmentation step for the defect area,which provides an effective method for defect recognition and measurement ofindustrial CT images with weak edge defects.