Research On Algorithms For Industrial Computed Tomographic Image Contour Processing And Vectorization

The internal structure of an object can be measured by industrial computedtomography (CT) systems non-destructively with non-contact，which has been widelyused in various fields. The parts’ reverse modeling or dimensional measurement basedon industrial CT testing has become one of the hot spots in industrial CT technologyresearch，and would be a supplement to the accurate measurement of complex parts inreverse engineering. The output of industrial CT system are cross-sectional imageswhich contain important geometry information of the target object including its shapeand size，and the objects’ dimensional drawings or three-dimensional CAD modelscould be reconstructed by these information. The cross-sectional contour shape featuresof most mechanical parts is a combination of these basic plane geometry primitives suchas straight line, circle, arc, ellipse and free curve etc. Image vectorization technique isclosely to restore these primitives’ geometric parameters and the correspondinginformation about position and size of the cross-sectional contour, providing importantdesign parameters for the model reconstruction. Therefore, the image processing andvectorization is the key for accurate model reconstruction in reverse engineering basedon industrial CT testing.Firstly, the image contour processing method was studied. image contourprocessing is required to go through image enhancement, contour extraction andcontour tracking steps in turn. In image enhancement, the histogram transform and theimage smoothing methods are mainly used to improve image’s quality. In contourextraction, two kinds of method, contour extraction method based on edge detection andmethod based on image threshold segmentation, are used to process to differentindustrial CT image contours; In contour tracing, using the Freeman chain code methodto track the contour could get an orderly sequence of contour points, and then store theinformation of these orderlypoints in a data linked list, preparing for the next stage ofimage contour vectorization.Secondly, an image contour vectorization method is researched. In this paper, animproved existence probability circle contour detection method is proposed to recognizecircle features in the first round, then these circle features are vectorized andcorresponding information of these contours are stored in the circle parameter linked list.This improved method could reduce the computation time, save memory space and enhance the efficiency of the circle feature recognition compared with the originalmethod. The second round is to recognize the rest of complex contours which are notrecognized as circle feature contours. Using the method based on the curvature of thecontour local curvature extreme points and the rough linear feature contour segmentsare extracted firstly. Then the improved set intersection method is used to recognize theremaining contour fragments into a series of connected line segments. From these linesegments, some sequences line segments are filtrated to construct circular arcs, usingthe perpendicular bisector method to recognize these circular arcs. At last, the adjacentline segments, which are approximately collinear, can be merged to a line.Finally, all the algorithms mentioned above would be programmed by using VisualC++6.0to develop the industrial CT image contour processing and vectorizationsystem software. This software can output the vectorization results of image contours inthe form of DXF files for CAD system softwares. Using this software, some industrialCT images were tested, the test results show that industrial CT images can effectivelyconvert into CAD vectorization images, the main feature information of image contouris restored as accurately as possible, and contours’ curve characteristic parameters canbe successfully obtained by means of the proposed industrial CT image contourprocessing and vectorization method.