Segmentation tool for images of cracks

TL;DR

A semi-automatic crack segmentation tool that eases the manual segmentation of cracks on images needed to create a training dataset for a machine learning algorithm and shows potential to be an adequate alternative to the manual data annotation.

Abstract

Safety-critical infrastructures, such as bridges, are periodically inspected to check for existing damage, such as fatigue cracks and corrosion, and to guarantee the safe use of the infrastructure. Visual inspection is the most frequent type of general inspection, despite the fact that its detection capability is rather limited, especially for fatigue cracks. Machine learning algorithms can be used for augmenting the capability of classical visual inspection of bridge structures, however, the implementation of such an algorithm requires a massive annotated training dataset, which is time-consuming to produce. This paper proposes a semi-automatic crack segmentation tool that eases the manual segmentation of cracks on images needed to create a training dataset for a machine learning algorithm. Also, it can be used to measure the geometry of the crack. This tool makes use of an image processing algorithm, which was initially developed for the analysis of vascular systems on retinal images. The algorithm relies on a multi-orientation wavelet transform, which is applied to the image to construct the so-called "orientation scores", i.e. a modified version of the image. Afterwards, the filtered orientation scores are used to formulate an optimal path problem that identifies the crack. The globally optimal path between manually selected crack endpoints is computed, using a state-of-the-art geometric tracking method. The pixel-wise segmentation is done afterwards using the obtained crack path. The proposed method outperforms fully automatic methods and shows potential to be an adequate alternative to the manual data annotation.

Figures (9)

• Figure 1: Example of a steel bridge structure image where common crack detection algorithms tend to have a high false call rate.
• Figure 2: Examples of the orientation scores: a) Orientation score of an image with lines; b) Orientation score of an image with a circle.
• Figure 3: Close up view of a crack with their calculated crack path retrieved by two methods: The red line shows the track obtained by tracking in $\mathbb{R}^2$ while the white line shows the spatially projected track obtained by tracking in $\mathbb{M}_2 = \mathbb{R}^2\times S^1$.
• Figure 4: a) Input image with chosen endpoints; b) Real part of the cake wavelets with angle $\theta$ equal to $0$ and $\frac{\pi}{2}$ respectively.
• Figure 5: a) Results of applying filters from figure Fig. \ref{['fig:cake wavelets']} to the image shown in Fig. \ref{['fig:input image']}; b) Orientation score of image shown in Fig. \ref{['fig:input image']}. $\theta$-levels represented by blue and red rectangles correspond to a).