Affiliations: [a] Centre de Recherche en Soudage et Contrôle, Algiers, Algeria | [b] Lab. Math. et leurs Interactions, Centre Univ. de Mila, Algeria | [c] Sp Lab., Université Constantine 1, Constantine, Algeria
Correspondence:
[*]
Corresponding author: Aicha Baya Goumeidane, Centre de Recherche en Soudage et Contrôle, Algiers 16002, Algeria. E-mail: [email protected]
Abstract: A perfect knowledge of a defect shape is determinant for the analysis step
in automatic radiographic inspection. Image segmentation is carried out on
radiographic images and extract defects indications. This paper deals with
weld defect delineation in radiographic images. The proposed method is based
on a new statistics-based explicit active contour. An association of local and
global modeling of the image pixels intensities is used to push the model to
the desired boundaries. Furthermore, other strategies are proposed to
accelerate its evolution and make the convergence speed depending only on
the defect size as selecting a band around the active contour curve. The
experimental results are very promising, since experiments on synthetic and
radiographic images show the ability of the proposed model to extract a
piece-wise homogenous object from very inhomogeneous background, even in a bad quality image.
Keywords: Welded joint, radiographic inspection, active contours, local statistics-based model, band region selection
