Unsupervised ridge detection using second order anisotropic Gaussian kernels

• We introduce the use of second order anisotropic Gaussian kernels for ridge detection.
• We present a dataset for ridge detection that contains 100 original in vitro fungi images.
• We analyze and test the use of multiscale anisotropic kernels in that same dataset.

We propose the use of the second derivative of Anisotropic Gaussian Kernels for ridge detection. Such kernels, which have proven successful in edge and corner detection, offer interesting advantages over isotropic kernels. In the case of ridge detection, these advantages include the increase of the sensitivity at junctions, as well as an improved characterization of blob-like artefacts. We do not only illustrate these advantages on synthetic images, but also perform a comparison on a new dataset for line detection, which is composed of 100 images of in vitro fungi.