A Novel Fuzzy C-Means Clustering Framework for Accurate Road Crack Detection: Incorporating Pixel Augmentation and Intensity Difference Features

Abstract

The widespread occurrence of fractures in roads globally threaten traffic safety and demand substantial annual maintenance costs. Expenses can be substantially reduced by detecting fractures promptly, however manual methods are less rapid and inaccurate. Although automatic crack detection offers efficiency, but challenges like low contrast and background noise in pictures can impact its accuracy. To address these obstacles, this study proposes a potent Fuzzy C-Means clustering technique to enhance automated fracture detection. This approach employs pixel augmentation through a scaling factor to improve pixel details by examining the ratios from individual to cumulative values. Additionally, it considers the sum of the total ratio value and the minimum-to-maximum intensity within a 3x3 window, prior to segmentation. Moreover, the method identifies intrinsic pixel connections through absolute intensity differences, supporting crack detection. It also effectively detects cracks from unfamiliar images across diverse scenarios, without the need for a training dataset. According to experimental results, an enhanced Fuzzy C-Means Clustering approach for road crack detection, achieving superior precision, recall, and F1 scores (86.68, 88.53, 87.59) compared to K-Means Clustering (76.82,78.05,77.43), Fuzzy C-Means Clustering (79.76,80.72,80.23), and Manhattan distance based fuzzy C-Means clustering (84.09, 86.14,85.10). Additionally, it also reduces iteration counts, ensuring computational efficiency. These results validate its robustness and effectiveness, making it a promising solution for automated road crack detection systems.