Abstract
The interactive image segmentation algorithm can provide an intelligent waysto understand the intention of user input. Many interactive methods have theproblem of that ask for large number of user input. To efficient produceintuitive segmentation under limited user input is important for industrialapplication. In this paper, we reveal a positive feedback system on imagesegmentation to show the pixels of self-learning. Two approaches, iterativerandom walks and boundary random walks, are proposed for segmentationpotential, which is the key step in feedback system. Experiment results onimage segmentation indicates that proposed algorithms can obtain more efficientinput to random walks. And higher segmentation performance can be obtained byapplying the iterative boundary random walks algorithm.