Detection of Curvilinear Objects in Noisy Image using Feature-Adapted Beamlet Transform
Sylvain Berlemont, Aaron Bensimon, Jean‐Christophe Olivo‐Marín · 2007
This paper addresses the problem of detecting features running along lines or piecewise constant curves. Our method is adapted either for common image features like edges or ridges as well as any kind of features that can be designed by a priori knowledge. The main contribution of this paper is to unify the well-known Beamlet transform, introduced by Donoho et al, with linear filtering technique in order to define what we call the feature-adapted Beamlet transform. If the desired feature is chosen to belong to the class of steerable filters, our method can be achieved in linear time and can be easily implemented on a parallel machine. We present some experimental results both on edge- and ridge-like features that demonstrate the substantial improvement over classical feature detectors.