Multimedia Tools and Applications, 21, 225–242, 2003 c  2003 Kluwer Academic Publishers. Manufactured in The Netherlands. An Efficient Active Contour Model Through Curvature Scale Space Filtering FARAHNAZ MOHANNA F.Mohanna@ee.surrey.ac.uk FARZIN MOKHTARIAN F.Mokhtarian@ee.surrey.ac.uk Centre for Vision, Speech, and Signal Processing, Department of Electronic and Electrical Engineering, University of Surrey, Guildford, England GU2 7XH, UK Abstract. Active contour models can be successfully used in multimedia database retrieval systems if they have good accuracy and high speed. The majority of existing active contour models do not lock on to interest objects very accurately and quickly especially in complex images. The behavior of the active contour is generally controlled by its internal and external energies. Internal energy is composed of two parts; the first part acts to shorten the active contour as it iterates towards the interest object, while the second part is the curvature of the active contour and forces smoothness of active contour during its movement towards interest object. In this paper, first a reformulated internal energy is proposed to improve the computation of curvature at point v i by making use of the three points v i -1 , v i and v i +1 . Second, an accurate and high speed active contour model, SAC is proposed based on reformulating internal energy by removing the curvature part and using Gaussian filtering with low scale of smoothing. The SAC model has only one parameter that affects the internal energy of active contour and as a result of using the Curvature Scale Space (CSS) 1 technique for smoothing, the SAC model is more independent of model parameter setting and the initial snake. Keywords: active contour, dynamic programming, curvature, energy-minimising, curvature scale space, edge detector 1. Introduction In recent years, active contour models or snakes have become one of the most powerful algorithms for image segmentation, boundary extraction, image matching and tracking. Our interest in active contour models comes from their use as a user interface for multi- media database retrieval using shape content. In such a shape-based multimedia database retrieval system, the user submits a query shape and then will expect the system to lo- cate all instances of similar shapes in database. An active contour can be successfully used for these tasks if it has good accuracy and high speed. A number of models have been proposed for active contours. The performance of active contours in these models depends on the proper setting of model parameters and the initial snake. Due to these shortcomings, majority of existing active contour models often fail to converge to the de- sired solution especially in complex images in which objects are very close to each other, see figure 1. The general idea of active contour models is as follows: first user places a closed contour around a target object in an image. Then the constraint forces act on ini- tial snake and push it towards object until it locks on to object as close as possible. The basic active contour model was proposed by Kass et al. [13]. Its algorithm has a num- ber of shortcomings which Amini et al. [2] pointed out and the solution was presented as a