Fully Automatic Model Creation for Object Localization utilizing the Generalized Hough Transform.
Heike Ruppertshofen, Cristian Lorenz, Peter Beyerlein, Zein Salah, Georg Rose, Hauke Schramm · Bildverarbeitung für die Medizin · 2010
An approach for automatic object localization in medical images utilizing an extended version of the generalized Hough transform (GHT) is presented. In our approach the shape model in the GHT is equipped with specific model point weights, which are used in the voting process. The weights are adjusted in a discriminative training procedure, which aims at a minimal localization error of the GHT. Furthermore, these weights yield information about the importance of points, such that non-discriminative points can be excluded from the model. Thereby, the size of the model is decreased, thus reducing processing time. The algorithm is presented in 2D but should be easily extendable for 3D images.