Defect segmentation of fiber splicing on an industrial robot system using GMM and graph cut

Abstract

A novel defect segmentation method, which utilizes both the Gaussian Mixture Model (GMM) and the Graph Cut Model (GCM), is presented to solve the defect segmentation problem of the hot image for the fiber splicing process on our industrial robot system. Since the fiber has a plastic surface, the LED lamp will create a highlight region in the fiber center when the camera collects the image data during the splicing process. Unfortunately, this highlight region always submerges the defect region. To solve this problem, both the image samples of normal mode and those of the defect mode are employed as the prior information to improve the segmentation performance. When implementing our method, first the GMM and the image samples of normal mode are used to build the statistic illumination model of the spliced fiber. The log histogram is tuned by the GMM components. Once the GMM is built, it can be utilized to restrain the highlight of the defect images. Then the GCM and the image samples of defect mode can be employed to segment the defect region and analyze their region features. Many simulation results have proved the effect of our proposed method.