Detection Of Defects In Castings Based On Thresholding Segmentation

Some castings can't meet the technical requirements because of the drawbacks of the manufacturing technique. On line inspection of the casting quality is very crucial. The technology that combining X-ray inspection system and digital image processing technology can detect the defects automatically, making the detection results objective, scientific and standard. In particular, the digital image processing method is one of the key parts in the X-ray inspection of castings. Considering the poor quality of the X-ray image of castings, including a low contrast and blurry edges, in this thesis, we propose two new image segmentation methods, which can be used to detect the typical internal defects in castings: air holes, shrinkage cavities and foreign objects.In terms of some prior knowledge, a bound histogram can be used to eliminate some interferences and elements that are not concerned, so it makes the image processing simple and feasible. Firstly, we propose a new image segmentation method based on bound histogram, called maximum fuzzy exponential entropy criterion. This method can overcome the drawbacks of Shannon entropy that are applied in image segmentation. Taking ambiguity or uncertainty into consideration in image segmentation, first of all, utilizing a parametric membership function to fuzzying image, then obtain the fuzzy 2-partion of object and background, as a result, an optimal threshold can be obtained by maximizing the fuzzy exponential entropy between the object and background.Secondly, in order to overcome the shortcomings in the maximum entropy criterion: many natural logarithmic operations are required. Considering the gray-level information and space neighborhood information at the same time, a maximal correlation criterion based on the two-dimensional bound histogram image thresholding algorithm is proposed in this thesis. The two-dimensional gray-level threshold may be determined by the maximizing the correlations associated with the distributions of the object and background classes. To improve the computational speed, a recursive algorithm is also presented, which can reduce re-calculations.Application of both methods to the detection of defects in casting, the results show that these both methods can correctly and quickly segment the defects in the X-ray image of castings with good robustness.