Image segmentation using a level set method based on local energy and gradient

To achieve quick and exact segmentations for intensity inhomogeneous images with complicated structures, we propose a novel level set method based on local region energy and gradient sensitivity, and initialize the level set automatically using the global intensity information. It is an improvement of minimization of region-scalable fitting energy with the introduction of two gradient sensitive energy functions. In this model, the local region energy based on the gray fitting functions acts as an external drive, which has an advantage in segmenting intensity inhomogeneous images. Gradient sensitive items can determine their drive directions of level sets adaptively according to image features. The external one draws zero level sets to the object boundaries, while the internal one pushes zero level sets out of intensity homogenous regions. Thereby, our method not only speeds up the convergence and improves robustness of level sets, but also copes with weak boundaries as the gradient sensitivity could be adjusted to respond to different gradient intensities. Meanwhile, interactive initialization is unnecessary. Experiment results demonstrate that our method has faster convergence, better robustness and good performance in segmenting intensity inhomogeneous images.