Study On Algorithms Of Geometric Primitive Recognition For Automatic Vectorization Of Industrial Computed Tomographic Image

With the development of Industrial Computed Tomography(ICT) and Reverse Engineering(RE), the measurement and CAD model reconstruction of workpieces which have complex closed internal features based on ICT system have been the relevant research hot topics. The 2-dimensional gray slicing image which is obtained by the ICT system have to execute vectorization for using it in CAD software; however, the traditional vectorization algorithms are easy to recognize an entire geometric primitive into a figure consisting of a series of short line segments for low recognition accuracy, which are difficult to fufill the requirement of practical application. This paper researchs automatic vectorization of industrial CT image for achieving the conversion from industrial CT image to 2-dimensional CAD graph, the algorithms of geometric primitive recognition for automatic vectorization of industrial CT Image are researched, and the automatic vectorization software of industrial CT image based on Visual C++6.0 is developed.Firstly, the binary image is obtained by executing image enhancement, median filter, thresholding or other preprocessing; then, the contour is extracted by the edge detection algorithm based on binary image, the acquired edge information is stored in the chained list. For this contour, the circle is recognized by an improved algorithm based on existence probability map, the ellipse is recognized by the randomized Hough transform algorithm of selecting 3 points(RHT₃), while the line is recognized with the set intersection algorithm of fitting a straight line, and the circular arc is recognized by means of the perpendicular bisector tracing algorithm. Finally, the acquired element parameters are stored in the circle list, elliptic list, linear list and arc list respectively, and the drawing exchange file(DXF) is produced to output the vectorization data.Improvement of existence probability map, the algorithm of recognizing ellipse and the development of vectorization software are researched especially in the thesis. To overcome the defect of excessive false peaks in original existence probability map, the Matlab is utilized to get the new existence probability map which directly reflects the actual locations of circles in industrial CT image; meanwhile, the efficiency of circle detection is improved. Adopting the algorithm of RHT₃ to recognize ellipse, the problem of useless sampling and accumulation due to direct randomized Hough transform is well solved, and the advantages of good anti-noise and strong robustness are retained. The developed vectorization software of industrial CT image comes with merits, such as easy operation, good stability, high fault tolerance and less human-computer interaction.Experiments are conducted with two real industrial CT images using the developed vectorization software, the results indicate that these algorithms are feasible and effectual, which have high accuracy; besides, the developed vectorization software can meet the requirement of converting industrial CT image into 2-dimensional CAD graph.