Adaptive Thresholding with Iterative Fuzzy Logic Based Image Enhancements for Wiper Arm Defect Detection

Quality control is an essential component in manufacturing process that ensures defective products are not shipped to customers. For a product like wiper arms, sophisticated methods for defect detection are required due to challenges of complex structures and reflective surfaces. This paper aimed to show enhancement to the previous solution by improving the image enhancement and thresholding steps, as well as the defect detection algorithm applied. In this proposed approach, the image is segmented using an improved Otsu-based threshold value selection to acquire the region of interest (ROI). The image enhancement step utilizes a modified iterative Fuzzy Clipped Contrast-Limited Adaptive Histogram Equalization (FC-CLAHE) that evaluates the clip limit required via contrast and entropy of images. This enhanced image is then binarized via adaptive thresholding with probabilistic Hough transform (PHT) and convex hull algorithm to preserve the car wiper arm features. The defect is then detected using a multistage blob detection algorithm that involves blob contour analysis based on contour area and contour mean values. Test results showed significant reduction in false negatives from 52% to 12.4%.

I. Introduction

Defect detection is a common process in manufacturing and is an important part in the quality control aspects of a product by regulating the product quality. In defect detection, human vision can be inconsistent and inaccurate due to fatigue. In order to overcome this drawback, defect detection through automated visual evaluation systems has been developed with suitable image processing techniques so as to provide accurate results that are repeatable, which is a superior solution that can replace human evaluation-based detection systems. In general, the defect detection architecture consists of the input image, the image pre¬processing block, the defect detection block and the output.