A multiscale image segmentation method. (April 2016)

• Record Type: Journal Article
• Authors: Li, Yafeng; Feng, Xiangchu
• Is Part Of: Pattern recognition. Volume 52(2016:Apr.)
• Page Start: 332
• Page End: 345
• Publication Date: 2016-04
• Subjects: Image segmentation -- Multiscale decomposition -- Variational model -- Total variation; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2015.10.004
• Languages: English
• ISSNs: 0031-3203

Abstract

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