Cold-rolled Strip Surface Defects Based On The Optimization Of The Filter, Split And Merge,

In the defect detection of the surface of the steel rolling armor plat, the computer vision show the importance more and more. Image segmentation is important in this system. The background of surface image of steel rolling armor plate is very complex, some of them are common armor plate image and galvanization armor plate image etc. By reason of absence of general image segmentation theory, it is difficult to find a proper image segmentation arithmetic that is effective both to common armor plate image and galvanization armor plate image. At the same time, most of the huge steel-rolling factory has been improving their requirement on automatic detecting system for surface quality of armor plate. So this paper designed new image segmentation arithmetic for the surface image of steel rolling armor plate based on digital filter. The major job and conclusion of this paper are:1. Analyzed the image collecting process and type of the noise. Designed Auto-adapted filter to improve the quality of the image.2. Designed three criterion functions. Based on these functions, designed three optimized filters and also showed the experiment results.3. Supervised and unsupervised defect detection is addressed here. Under the supervised defect detection, analyzed characteristic of these three filters. At the same time, compared their shortage and advantage. In corresponding to the shortage of optimized filter, designed multi-channel optimized filter. The experiments show that the optimized filter and multi-channel optimized filter are both robust and can segment the image finally. But the result is affected by the filter's dimension.4. In order to realize the unsupervised defect detection, this paper designed the method of multi-channel filter merge. The experiments show the feasible and reliable of this method.5. So as to get integrated defect from the filter image, this paper addressed the method to integrate the interlinked defects and not completed defect. The experiments show the effective of this method.