SHAPE PRIOR FOR AN EDGE-BASED ACTIVE CONTOURS USING PHASE CORRELATION Mohamed Amine Mezghich, Malek Sellami, Slim M’Hiri and Faouzi Ghorbel GRIFT Research Group, CRISTAL Laboratory ´ Ecole Nationale des Sciences de l’Informatique, ENSI Campus Universitaire de la Manouba, 2010 Manouba, Tunisia ma.mezghich@cristal.rnu.tn, {malek.sellami,slim.mhiri,faouzi.ghorbel}@ensi.rnu.tn ABSTRACT In this paper, we intend to present a new method to incorpo- rate geometric shape prior into an edge-based active contours in order to improve its robustness to partial occlusions, low contrast and noise. The proposed shape prior is defined after the registration, based on phase correlation, of binary images associated with level set functions of the active contour and a reference shape making the model invariant with respect to Euclidean transformations. Experimental results on synthetic and real images show the ability of the proposed approach to constrain an evolving curve towards a target shapes that may be occluded and cluttered under rigid transformations. Index Terms— Active contours, shape prior, phase cor- relation, rigid transformations 1. INTRODUCTION Segmentation is a crucial step in image processing. Its role is to partition the image into homogeneous regions. Active contours [1, 2, 3, 4, 5] have been widely used in this field. One can classify them into two families : The boundary- based approach which depends on an edge stopping func- tion to detect objects and the region-based approach which is based on minimizing an energy’s functionnal to segment ob- jects in the image. Region-based models are more robust than boundary-based methods since the information of the whole region is explored. However, they need more computations and the region is restricted to be uniform and occlusion free. Given that classical active contours are intensity-based mod- els, there is still no way to caracterize the global shape of an object. Especially in presence of occlusions and clutter, all the previous models converge to the wrong contours. To solve the above mentioned problems, different attempts incorporate shape prior into the active contour models. Leventon et al., [6] associated a statistical shape model to the geodesic active contours [3]. At each step of the surface evolution, the max- imum a posteriori position and shape are estimated and used to move globally the surface while local evolution is based on image gradient and curvature. Chen et al., [7] defined an energy’s functional based on the quadratic distance between the evolving curve and the average shapes of the target ob- ject after alignment. This energy is then incorporated into the geodesic active contours. Bresson et al., [8] extended [7] ap- proach by integrating the statistical model of shape proposed by [6] in the energy functional. Fang and Chan [9] introduced a statistical shape prior into the geodesic active contour to de- tect partially occluded object. To speed up the algorithm, an explicit alignment of the shape prior model and the current evolving curve is done to calculate pose parameters. Foulon- neau et al., [10] introduced a geometric shape prior into a region-based active contours [5] based on the Legendre mo- ments of the characteristic function and in [11], Charmi et al., defined a geometric shape prior for the region-based active contours after alignment of the evolving contour and the ref- erence shape. It’s well known that shape priors based on con- tour alignment methods force these approaches to segment only single object in the image and go without the contribu- tion of level set, i.e. its ability to segment multiple objects at once (see [8]). Besides, contour alignment methods are not adapted to estimate the rigid transformation parameters in the case of objects with holes (which often occurs in medical im- agery like MRI brain’s white matter). This justifies the use of the registration methods instead of those based on contours alignment. In this work, we focus on adding a new geomet- ric shape prior to an edge-based active contours [4] that use the relative motion parameters between objects of the same shape and have different pose, size and orientation, estimated by phase correlation. We assume that the shape of reference is known in advance (like the work of [10]). The improved model can retain all the advantages of the level set approach and have the additional ability of being able to handle the case of images with multiple objects in presence of noise, partial occlusions and low contrast. The remainder of this paper is organized as follows : In Section 2, we will recall the used shape registration method based on phase correlation. Then, the proposed shape prior will be presented in Section 3. Ex- periments will be presented and commented in order to study the robustness of the model in Section 4. Finally, we conclude the work and highlight some perspectives in Section 5. EUSIPCO 2013 1569746723 1