A survey on curvilinear object segmentation in multiple applications. (December 2016)

• Record Type: Journal Article
• Authors: Bibiloni, P.; González-Hidalgo, M.; Massanet, S.
• Is Part Of: Pattern recognition. Volume 60(2016:Dec.)
• Page Start: 949
• Page End: 970
• Publication Date: 2016-12
• Subjects: Image processing -- Pattern recognition -- Curvilinear objects -- Segmentation -- Survey -- Review; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2016.07.023
• Languages: English
• ISSNs: 0031-3203

Abstract

Abstract: Curvilinear object segmentation is a paramount step for many applications ranging from medical to aerial image processing. In particular, vessel segmentation in retinal images, detection of spiculated lesions in mammograms or extraction of airways in CT scans provide essential information to experts to evaluate, diagnose and propose a treatment. The significance of these applications has conducted important efforts to propose curvilinear object segmentation algorithms based on the most different techniques. The main objective of this review is to clearly present the similarities and differences between curvilinear structures in different applications and the different techniques used to segment them more effectively. To do so, we propose a general definition of curvilinear structures that encompasses the distinct models considered in the literature. In addition, we analyse and classify the mathematical techniques used to segment the curvilinear structures found across all considered applications, studying their strengths and weaknesses. In particular, we present the most relevant benchmarks related to curvilinear object segmentation as well as the best algorithms according to several performance measures. By doing so, it is acquired a wider point of view to extend the results from some fields to others, and to understand under which conditions some methodologies should be favoured over the rest of them. Abstract : Highlights: We study what curvilinear objects are found across different imagery techniques. We study and classify which algorithms have been used to segment them. We present multiple performance measures of such segmentation algorithms. We discuss which approaches are better to segment which curvilinear structures.