Target region preserving smoothing for selective image segmentation
Wenxiu Zhao, Xiaofang Li, Chunyu Yang, Leigang Huo · Research Square · 2022
Abstract Selective image segmentation is an important and challenging task, which aims to segment a subset of target objects or regions of interests in an image. In this paper, we propose a two-stage selective segmentation method. In the first stage, we propose a new smoothing model, which aims to reduce the influence of noise or cluttered background on segmentation. The model consists of a regular term and a fidelity term. First, we define a plug-and-play regular term by incorporating the trained fractional optimal control network. Then we introduce the fidelity term a weight dependent on the distance function of a set of marker points in the target region. In addition, the proposed model has a closed-form solution, which is easy to implement. In the second stage, the smooth image obtained in the first stage is divided into background region and target region with a simple threshold value. Extensive experiments show that our smoothing model can greatly facilitate the second phase, and our method performs better than some existing related methods in terms of either visual assessment or quantitative evaluation.