A multiscale image segmentation method

摘要: This paper presents a novel image segmentation framework that combines and feature extraction into unified model. The proposed model consists of two parts: the part multiscale decomposition part. In model, relies on intensities in regions interest while depends features different scales. facilitates process since region can be easily detected from proper scale. total variation projection regularization (TVPR) is used to preserve geometric shape segmented regions. According significance TVPR parameters, an adaptive parameters selection method presented edges well preserved. able deal with intensity inhomogeneities mixed noises often occurred real-world images, which present challenges segmentation. Numerical examples synthetic real images are given demonstrate effectiveness method. proposes framework.A within our framework.Total model.We for segmentation.The experimental results show