Skeleton Generation Technique Of Casting Defect Based On Image Interpret Process

With the performance of computer hardware has been improved, the development of machine vision technology and the demand of modern casting industry, using automatic detection algorithm to assist manual or even substitute manual for quality supervision in casting production will become a widely trend. In the process of tuning automatic defect detection algorithm, it often requires a certain amount of test image of casting defect. Due to the progress of the casting process,the incidence of defect which is obtained by production collection is greatly reduced. In order to provide more sample images in a short time, related research of defect simulation technique has been studied. Among them, a sort of algorithm using skeleton as input primitives to generate defect contour templates. In their experiments, a source of skeleton sample is extracted from real defects by using extract algorithm, and another source is the way to drawing by hands. Using the algorithm to extract skeleton, firstly it needs to consider the robustness of extraction method, by using some of the extraction method based on boundary information, it may involves the image segmentation problem.When using hand-drawn ways with large sample demands, it requires a lot of redundant works and it is inefficient.This paper will regard the skeleton of the dendritic shrinkage defects as the object of the research, and proposes a defect skeleton extraction technique based on gradient vector flow diffusion.The proposed method can exactly locate the position of defect skeleton and avoid the binary process of segmenting the defect target.Aiming at a sort of defect simulation technique which using hand-drawn skeleton as a input primitive,this paper builds a mode of skeleton keyframe base on the concept of joint point and proposes a method to generate skeleton sample automatically base on the simulative growth process of defect.This paper uses the generated skeleton samples to conduct related defect simulation experiments. It proposes using the fuzzy numerical descriptors to describe the features of defect simulation image. Combining subjective and objective ways, the similarity between defect simulation image and real defect image was evaluated.