Semi-automatic spline fitting of planar curvilinear profiles in digital images using the Hough transform. (February 2018)

• Record Type: Journal Article
• Authors: Conti, Costanza; Romani, Lucia; Schenone, Daniela
• Is Part Of: Pattern recognition. Volume 74(2018:Feb.)
• Page Start: 64
• Page End: 76
• Publication Date: 2018-02
• Subjects: Hough transform -- Profile recognition -- Spline fitting -- G1-continuity -- Cusps; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2017.09.017
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: Spline-based algorithm for the semi-automated recognition of planar curvilinear profiles in digital images. Neither a family of predefined curves nor a look-up table of prototypal shapes required as input. Robustness with respect to background noise guaranteed. Favorable comparison with existing profile extraction algorithms relying on the Hough transform technique. Effectiveness tested on real medical images. Abstract: We develop a novel method for the recognition of curvilinear profiles in digital images. The proposed method, semi-automatic for both closed and open planar profiles, essentially consists of a preprocessing step exploiting an edge detection algorithm, and a main step involving the Hough transform technique. In the preprocessing step, a Canny edge detection algorithm is applied in order to obtain a reduced point set describing the profile curve to be reconstructed. Also, to identify in the profile possible sharp points like cusps, we additionally use an algorithm to find the approximated tangent vector of every edge point. In the subsequent main step, we then use a piecewisely defined Hough transform to locally recognize from the point set a low-degree piecewise polynomial curve. The final outcome of the algorithm is thus a spline curve approximating the underlined profile image. The output curve consists of polynomial pieces connected G 1 continuously, except in correspondence of the identified cusps, where the order of continuity is only C 0, as expected. To illustrate effectiveness and efficiency of the new profile detection technique we present several numerical results dealing with detection of open and closed profiles in images of different type, i.e., medical and photographic images.