Abstract
Automatic detection of complex cracks on rough concrete surfaces via image processing is a challenging task. The most current effective methods involve deep learning schemes. These are usually computationally and structurally complex. Recently, relatively simplified algorithms were developed for effective segmentation of crack features. However, these approaches still could not consistently and accurately extract such features from extremely noisy images of rough concrete surfaces with complex crack patterns. This study describes crack feature segmentation algorithms based on wavelet coefficient adjustment, nonlinear filter pre-processing, saturation channel extraction, adaptive threshold-based edge detection and fuzzy clustering-based area classification. Additional modifications include a new energy function for active contour segmentation algorithm. Adaptive localized mask generation is also proposed for automatic region-based segmentation. Furthermore, a binary fusion stage is incorporated for improved edge feature extraction. The quantitative and visual evaluation of the proposed schemes show improvement in results compared to several recent state-of-the-art algorithms.