Variational selective segmentation model for intensity inhomogeneous image

Variational selective image segmentation models aim to extract a particular object in an image depending on a set of user-defined prior points. The current model suffers from high computational costs due to the traditional total variation function that results in a slow segmenting process. In addition, it is not designed to segment images with intensity inhomogeneities. In this research, we formulate a new variational selective image segmentation model based on the Gaussian function. A Gaussian function is proposed to replace the traditional total variation function to regularize the variational level set function. To segment images with intensity inhomogeneities, the local image fitting idea was in corporate into the formulation. The efficiency of the proposed model was then assessed by recording the computation time while the accuracy was measured using Jaccard and Dice similarity values. Numerical experiments using synthetic, natural, and medical images demonstrate that the proposed model is about 6 times faster than the existing model, while the Jaccard and Dice values are about 11% and 7% higher, respectively, compared to the existing model. In the future, this research can be extended further into a 3-dimensional modeling and vector-valued image framework.