Improved Selective Segmentation Model Using One Level-Set

Abstract

Variational segmentation models have proven to be extremely efficient for segmenting and tracking boundaries and in most cases all of the boundaries in an image. Such models are for global segmentation. For a large class of image segmentation tasks where only one object is required to be extracted automatically, global models cannot deliver the solution and we need selective segmentation techniques. In this paper, we propose a novel, variational and single level-set function for the selective segmentation task. The model is much faster to implement than the previously dual level set model by Rada-Chen [29] by having the same efficiency and reliability. In comparing with interactive image segmentation algorithm of Nguyen-Cai-Zhang-Zheng method [2], our model shows some improvement in some cases. Several new ideas are incorporated in this new work: i) the distance function is only needed optionally and its inclusion does not affect the result; ii) an adaptive parameter is introduced in the edge detection function; iii) a new area-based fitting term is added to enhance the model's reliability (different from the idea of minimizing the area of the object). We develop an additive operator splitting method for solving the resulting Euler-Lagrange equation. Test results show that the new model finds the desired local boundaries successfully in various challenging cases and indeed it is not much dependent on the prior information of markers or the distance function based on them. More importantly, the new model gives an overall improvement over the previous models and can be recommended for selective segmentation.