A Hybrid Framework for Concrete Crack Assessment Using Grab-Cut and Improved Sobel Filtering

Cracks are irregularities in the physical form and scale of concrete constructions. Timely analysis of cracks is crucial to ensure structural health and safety. Crack assessments are traditionally done visually, non-destructively, and manually. These methods involve a significant amount of human engagement, as well as a skilled and qualified inspector. Furthermore, these procedures are biased and lengthy, usually missing to correctly identify cracks in complicated concrete structures. In response to these challenges, a hybrid approach that combines a grab-cut algorithm and an improved Sobel filter has been proposed for automatic crack recognition in concrete samples. The proposed method utilizes the image processing principle. The proposed method operates as a two-step model, utilizing a grab-cut algorithm to extract the cracks’ foreground and background regions. This advancement helps to mitigate challenges like noise and image quality. The next stage involves processing the segmented data using enhanced Soble filtering to accentuate the identified crack edges. To refine the efficacy of improved Soble filtering, the horizontal and vertical masks are advanced with the variance of the image. This modification provides a precise detection of edges as compared to a conventional mask. To judge the performance of the proposed method, images of self-prepared concrete specimens are used. We assess the performance both qualitatively and quantitatively, comparing it to established methodologies like Otu’s thresholding, Canny, Sobel, Log detector, Zero Cross detector, Robert, and Prewitt. The comparative quantitative study has been presented in terms of global consistency error, variation of information, Rand index, average Hausdorff distance, boundary F1 score, Jaccard index, and Dice coefficient. The following values are obtained: 0.2480, 0.8179, 0.9827, 25.5132, 0.9805, 0.9858, and 0.9878, respectively. The analysis exhibits that the proposed method surpasses the outcomes of the other pre-existing methods. Apart from this, a MATLAB-based graphical user interface for the proposed method has also been developed. The civil infrastructure maintenance agency and construction engineers can use the GUI for their structural maintenance tasks.