Abstract—This paper is devoted to present and discuss a model that allows a local segmentation by using statistical information of a given image. It is based on Chan-Vese model, curve evolution, partial differential equations and binary level sets method. The proposed model uses the piecewise constant approximation of Chan-Vese model to compute Signed Pressure Force (SPF) function, this one attracts the curve to the true object(s)’s boundaries. The implemented model is used to extract weld defects from weld radiographic images in the aim to calculate the perimeter and surfaces of those weld defects; encouraged resultants are obtained on synthetic and real radiographic images. Keywords—Active contour, Chan-Vese Model, local segmentation, weld radiographic images. I. INTRODUCTION OWADAYS the visual information has being introduced in very large applications, thank to that image processing posses more and more a crucial importance. Many axes had being created to recover all the problems and difficulties related to use images as input for an automatic system. One of those axes is the segmentation with which this present work is concerned. One of the applications of computer vision is devoted to Non Destructive Testing NDT by radiographic technique. In welding, industrial radiographic operation is similar to the medical one, it consists to submit a gamma rays or x-rays from its source through the welded join. The differences of the densities between the based material, the welded joint and defects are reflected on the radiographic films. The objective of our team is to segment those digital images in order to give them the structural forms for ulterior processing, such as computing the surfaces and the perimeters of weld defects with the aim to use them in NDT task. Segmenting images by deformable models or variationnal methods has known great success and wide using. Many functionals have being proposed. The classification of those models is variable according to on which we are based to do that. Two famous categories are often met in literatures; the first one is based on the terms that link the model to the image: it can be oriented edge or region. The second one is based on Y. Boutiche and N. Ramou are with Image and Signal Processing Laboratory, National Research Center on Welding and Control, CSC, Route de Dely Brahim B.P.64 , Algiers, Algeria, (phone: 213-21361850; fax: 213- 21361850 boutiche_y@yahoo.fr). M. Ben Gharsallah is with Research CEREP Unit, ESSTT, 5 Av. Taha Hussein, 1008, Tunis, Tunisie. the way to represent the curves: explicit representation or implicit one [1][2]. Almost all edge-based models use the gradient of the image ݑ ଴ to locate the objects’ edges. Therefore, the curve is locally stopping when it reaches high image gradients [1] [2], [3]. For that an edge-function is often used, which is strictly positive inside homogeneous regions and near zero on the edges, it is formulated as follow: ሺ|׏ ݑ ଴ |ሻ ൌ ଵ ଵା|׏ሺ ഑ כ௨ బ ሻ| ೛ , ݌ ൌ 1,2 (1) The gradient operator is well adapted to a certain class of problems like robustness to region inhomogeneities. They also have important drawbacks that make them inefficient on noisy images [4], very sensitive to initial conditions (when the contour initialization is not completely inside or outside the region to segment). Moreover, they are only able to segment regions with sharp edges, so this can result in failure when the region edges are smoother. Many works are focusing in overcoming those problem, for example in [5], the authors managed to improve the initialization problem, by creating a vector flow driving the active contour to high image gradients, but the sensitivity to noise still remains. On the contrary, the based-region approaches avoid the derivatives of the image intensity and they use statistical information of the image intensity to attract the curve evolution at the objects’ boundaries. Often we use the average intensities and standard deviation. However based-region approaches are more robust to the noises, it detects objects whose boundaries cannot be defined or are badly defined through the gradient, they automatically detect interior contours, the initialization could be anywhere on the image domain not necessary surrounded the objects, in addition, they have better tendency to compute a global minim of the functional [6][7]. For some specific applications, as the weld defects extraction, a global segmentation isn’t wanted, in the same time we want to benefit from all the advantages related to use region-based Models. For that and inspired from some recent works we propose an algorithm that allows a local segmentation by using statistic image’s information. This paper is organized as follows: after this section we recall the two models devoted to segment images by variational techniques. In section III we explain the proposed model Local Chan-Vese, which allows a local segmentation by using statistical image information. Section V is dedicated to the implementation in which we introduce the Binary Level Set, and the algorithm implemented during this work. The An Implicit Region-Based Deformable Model with Local Segmentation Applied To Weld Defects Extraction Y. Boutiche, N. Ramou, and M. Ben Gharsallah N World Academy of Science, Engineering and Technology International Journal of Electronics and Communication Engineering Vol:6, No:11, 2012 1267 International Scholarly and Scientific Research & Innovation 6(11) 2012 ISNI:0000000091950263 Open Science Index, Electronics and Communication Engineering Vol:6, No:11, 2012 publications.waset.org/10889/pdf