A local histogram based Chan-Vese model for segmentation

The Chan-Vese model is a successful Variational segmentation model for piecewise constant and smooth images. However, it is failed to detect tigers or leopards in an image. In this paper, we propose a modified model that segments tiger into a whole object. Our energy functional uses an improved fitting term based on local histogram and the energy minimizing is therefore looking for homogeneous histograms to distinguish objects. We present a numerical algorithm to compute the corresponding partial differential equation. The proposed model yields excellent performance on synthetic images and images from the Berkeley segmentation data set 500.