Learning to tune level set methods

Level set methods are very useful models in image segmentation, but require delicate adjustments of many parameters, which are typically determined empirically. This paper proposes a novel automatic method to address the challenge of parameter tuning for level set methods. It analyses the energy impact on the objects of interest during construction of the final contours, using a supervised machine learning approach. The method allows level set methods to automatically choose optimal values of the model parameters and extract objects at different granularities based on the training data. Experimental results demonstrate the capability of the method for accurate parameter tuning.