A variational model with hybrid images data fitting energies for segmentation of images with intensity inhomogeneity. (March 2016)

• Record Type: Journal Article
• Authors: Ali, Haider; Badshah, Noor; Chen, Ke; Khan, Gulzar Ali
• Is Part Of: Pattern recognition. Volume 51(2016:Mar.)
• Page Start: 27
• Page End: 42
• Publication Date: 2016-03
• Subjects: Image segmentation -- Calculus of variations -- Level set method -- Partial differential equations -- Edges -- Objects; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2015.08.022
• Languages: English
• ISSNs: 0031-3203

Abstract

Abstract: Level set functions based variational image segmentation models provide reliable methods to capture boundaries of objects/regions in a given image, provided that the underlying intensity has homogeneity. The case of images with essentially piecewise constant intensities is satisfactorily dealt with in the well-known work of Chan–Vese (2001) and its many variants. However for images with intensity inhomogeneity or multiphases within the foreground of objects, such models become inadequate because the detected edges and even phases do not represent objects and are hence not meaningful. To deal with such problems, in this paper, we have proposed a new variational model with two fitting terms based on regions and edges enhanced quantities respectively from multiplicative and difference images. Tests and comparisons will show that our new model outperforms two previous models. Both synthetic and real life images are used to illustrate the reliability and advantages of our new model. Abstract : Graphical abstract: Abstract : Highlights: We model image segmentation problem using product and difference images. The convex formulation of the model is also presented. The proposed model takes care of the minute details and inhomogeneity in an image. This model is compared with LCV and FGM of the active contour model. The competing energies are qualitatively tested by using Jaccord Similarity Index.