Automatic Welding Defect Detection of X-ray Images by Using Adaptively Regularized Kernel Fuzzy Technique Integrated with Edge-Based Level Set Function

Industrial images comprise complex configurations and their accurate segmentation is crucial for facilitating the delineation, characterization, and extraction of the region of interest. The edge-based level set (ELS) approach is one of the most often used in this field, but its main problem is the sensitivity to the initial position contour. In this work, we propose a hybrid image segmentation model using adaptively regularized kernel fuzzy technique (ARKF) integrated with edge-based level set function to solve this problem and enable welding defect detection. More specifically, our ARKF-ELS model comprises three key stages. The first stage applies the kernel fuzzy technique to isolate the cluster containing welding defects (regions of interest (ROIs)) from input image. In the second stage, this cluster is used to initialize the ELS method. In the third stage, the ARKF-ELS model is adopted to extract the weld defects. Experimental results on X-ray images demonstrate that the ARKF-ELS model can effectively extract regions of interest (ROIs) and confirm its efficiency in welding defects segmentation.