The multichannel integration active contour framework for crack detection

Xiaomin Xie,Kui Li,Tingting Wang,Yilin Xia,Bo Liu

doi:10.1177/1729881419852853

Xiaomin Xie, Kui Li + Show 3 more

Open Access

https://doi.org/10.1177/1729881419852853

Copy DOI

Abstract

Crack is one of the most important defects to evaluate the health of concrete buildings. Hence, accurate detection is of great significance for the infrastructure maintenance. In this article, an efficient multichannel active contour model for crack extraction is proposed, which integrates various features of the cracks. Firstly, the nonlocal means technique is adopted to eliminate the effects of noise while preserving the edge details. Then, the novel multichannel active contour model energy function is constructed, which considers three characteristics of the cracks: (a) the intensity features map, which is on the basis of the distinct intensity of the cracks; (b) the saliency feature map, which is obtained by the frequency-tuned salient region detection; and (c) the line-like feature map, which is enhanced by the multi-scale Hessian filtering. Also, the line-like feature map is binarized by a set of morphological operations and the Otsu thresholding to initialize the active contour. The proposed approach has been compared with the existing detection models on the public database and the real-world cracks. The experimental results show the effectiveness and efficiency of the proposed model.

Highlights

Concrete architectures are the most common infrastructures in cities
Three characteristics of the cracks assist to establish the energy function of the proposed multichannel active contour models (ACMs): (a) the intensity features map, which is based on the distinct intensity of the crack; (b) the saliency feature map, which is obtained by the frequency-tuned salient region detection; and (c) the linelike feature map, which is emphasized by the multi-scale Hessian filtering
The second part concentrates on the multichannel ACM which establishes the integration region-scalable fitting energy over three types of maps, that is, the intensity-based map, the visual salience map, and the multi-scale line-like feature map

Summary

Introduction

Due to the low tensile strength of the concrete architectures and the aging of buildings, various concrete diseases are easy to occur, resulting in engineering accidents, and even causing huge economic losses and personal injuries Among these defects, cracks are the early diseases of the concrete structures. For the training of the deep convolutional neural networks, a large number of annotated crack images are required, which restrict their real application It is difficult for a single detection algorithm to achieve ideal results. The energy functions of the active contour models (ACMs)[10,11,27] are established based on the characteristics of the cracks (intensity, edges, textures, etc.), which could be regarded as general and flexible frameworks. The third section presents the experiments and analysis; the conclusion is drawn in the fourth section

Methodology

H ð’Þdxdy ð þ ð 1 N

Conclusion