Research On The Segmentation Of Target Regions For Strip Surface Defects Image

The strip surface defects segmentation is a very important item for strip surface defects detection. Efficiency and accuracy of the defects segmentation will directly affect the classification and recognition, even affect the whole performance of detection system. Due to the variety of strip surface defects, the uneven illumination of background, the irregular of texture background, it is hard to find an effective and general method for the strip surface defect segmentation. Therefore, this paper in-depth researched the methods of strip surface defect segmentation, and proposed the strip surface defect segmentation method based on the convex active contour model.The main research contents and results are listed as follows:(1) By an overview of the basic requirements of strip surface defect online testing system, a basic processes of strip-line surface defects detection system is given, the system is divided into two parts:"real-time processing" and "on time processing". In order to ensure a low missing rate and false positive rate, and to meet the requirements of real-time detection system, this paper combined a simple method for rapid detection and an accurate target regions segmentation method to meet requirements of the real-time detection system. Namely, it used simple image gray projection method for rapid detection to meet the requirements of real-time detection system, while using a more accurate model based on active contour target region segmentation to ensure the lower missing rate and false positive rate.(2) By reviewing the Chan-V model, LBF models and several typical active contour model, an effective improvement program based on convex active contour model (CACM) is proposed. The model utilizes convex optimization technology which made a non-convex problem to a convex optimization problem, and also used the Split Bregman method. The model effectively solved the issues of the initial contour position sensitive of Chan-V model and the LBF model. At the same time, this model also used the local information, and achieved good effects in gray non-uniformity of the strip surface defect image segmentation. As the use of the Split Bregman solution method, the running speed has greatly improved than the other models. It can not only meet the rate requirements of the strip surface defect target regions segmentation, but also strive more time for the subsequent feature extraction and selection, and classification.(3) By reviewing and summarizing some typical multiphase active contour model for image segmentation, Vese-C model based on multi-target was used to segment multi-target regions of multi-phase defect image. The proposed CACM model was used to segment multi-target regions of two-phase defect image. Four categories common defects are experimented, which are inclusion, scratches, pitting, and wrinkles. Experimental results show that the defects target region have been accurately divided out, and it also will meet speed requirements of the strip surface defect target regions segmentation.