An affine shape constraint for geometric active contours - Emerging Trends in Image Processing, Computer Vision, and Pattern Recognition - page 418

CHAPTER 33
An affine shape constraint for
geometric active contours
Mohamed Amine Mezghich; Maweheb Saidani; Slim M'Hiri; Faouzi Ghorbel GRIFT Research Group, CRISTAL Laboratory,
Ecole Nationale des Sciences de l'Informatique (ENSI), Campus Universitaire de la Manouba, Manouba, Tunisia
Abstract
In this paper we intend to present an original approach that incorporates shape prior into a geometric
active contours. A Fourier-based shape alignment method is used to define prior knowledge from a giv-
en reference shape. At a first step, we propose an invariant shape prior with respect to some Euclidean
transformations (translation, rotation, and scaling factor). Then, we generalize to the case of affine ones.
In order to take into account the multireferences situation, we present a set of complete and stable invari-
ant shape descriptors computed using Fourier transform of the contours to choose the most suitable ref-
erence according to the evolving contour. To balance the strength between the image-derived force and
the proposed shape prior force, a weighting schema is introduced to guide the model evolution under
both forces. Experimental results show the ability of the new added term to constrain an evolving curve
meet target shape and show the new method benefits on segmentation results in presence of occlusion,
clutter, and noise.
Keywords
Active contours
Shape prior
Contours alignment
Fourier transform
Motion estimation
1 Introduction
Precise object of interest detection is of high need in many applications of image processing
such as medical imaging, tracking of moving object, and 3D object reconstruction. There have
been several works in this field of image segmentation and object detection. Among them we
can distinguish the classical active contours which are intensity-based models [ 1 - 5 ] and the
constrained models by a prior knowledge that are driven both by global geometrical [ 6 - 9 ] or
statistical [ 10 - 14 ] shape information and a local image information like gradient or curvature.