Detecting cracks in underwater concrete structures: an unsupervised learning approach based on local feature clustering

This paper presents an unsupervised approach for crack detection in underwater concrete structures. It is based on local feature clustering using K-Medians on Haralick texture features. An additional step for outliers removal is introduced, based on a bimodal Gaussian distribution for candidate blocks. This approach has been successfully tested on a dataset of 490 images, with quantitative results produced for a subset of 40 random images in the dataset.