Documente Academic
Documente Profesional
Documente Cultură
Web Site: www.ijettcs.org Email: editor@ijettcs.org, editorijettcs@gmail.com Volume 2, Issue 2, March April 2013 ISSN 2278-6856
Evolution of Contours in Deformable Models using Single and Grid Mask Initialization
Neeru Gulati
Department of Computer Science & Engineering Ambala College of Engineering & Applied Research, Devsthali, Ambala
Keywords: Active contours, energy minimization, segmentation, energy function, deformable models.
1. INTRODUCTION
Image segmentation is the part of every image processing application. Segmentation partitions an image into different parts or objects. It refers to the process of partitioning a digital image into multiple segments i.e. set of pixels, pixels in a region are similar according to some homogeneity criteria such as colour, intensity or texture, so as to locate and identify objects and boundaries in an image [1]. More precisely, image segmentation is the process of assigning a label to every pixel in an image such that pixels with the same label share certain visual characteristics. We can segment the image by using a PDE based method & solving the PDE equation by a numerical scheme. Image segmentation based on PDEs is mainly carried out by active contour model or snakes. This method was first introduced by Kass et al in 1987 [2]. Kass developed this method to find familiar objects in presence of noise and other ambiguities. . Snakes lock on to nearby minima in the potential energy generated by processing an image. The name deformable models or snakes first appeared in work by Terzopoulos and his collaborators. Active contours or snakes are computer generated curves that move within the image to find object boundaries under the influence of forces of curve and image itself. Active contours detect the objects in a given image, based on techniques of curve evolution, Volume 2, Issue 2 March April 2013
MumfordShah functional for segmentation and level sets. A formulation, to implement these active contours, was proposed by the level set theory [3]. One may visualize the active contour as a rubber band of arbitrary shape that is deforming with time trying to get as close as possible to the object contour. This model can detect objects whose boundaries are not necessarily defined by gradient, Active contours without edges, Active contour without edges for vector image, Active contours with multi-phases. Each one is having different speed and quality. All these methods require proper initialization of mask to evolve. Proper initialization of mask directly affects the performance and quality of segmentation. Different initial conditions can produces different results. This paper contains the discussion regarding the initialization of mask in any image followed by the segmentation using Chan-Vese based on curve evolution, Mumford Shah functional and level set [4]. In section II of this paper, Chan-Vese technique of active contours has been described, section III describes the method of initialization of mask as single and grid, the rest of the paper contains the comparison between the initialization as single mask and grid mask on different images of dental application.
2. IMAGE SEGMENTATION
Segmentation is the process of partitioning the image into set of distinct regions which are different in some important quantitative or qualitative way. Image segmentation based on PDEs is mainly carried out by active contour model or snakes. This method was first introduced by Kass et al in 1987 [2]. The original idea by Kass, Witkin and Terzopoulos [4] is to initialize a curve in an image and let this curve move until it adapts to the contour of a searched object.The motion of the curve is driven by the image itself. The driving force is obtained by defining a potential in the image that shall be small near objects contours. The objective is to minimize an energy which can be seen as a particular case of the minimal partition problem. But the classical snakes could detect the object with edges defined by gradient. In the level set formulation, the problem becomes a meancurvature flow-like evolving the active contour, which will stop on the desired boundary. Chan-Vese Energy function [4] is given by the following equation: Page 285
(1)
, , , are the parameters whose values lie between 0 and 1.The objective is to minimize the E, when E will be minimum C= C and image will be segmented. In ChanVese approach involves geometric active contour model (based upon Mumford Shah Functional) [4]. The model begins with a contour in the image plane defining an initial segmentation and then contour is evolved according to evolution equation. The goal of algorithm is to minimize this fitting energy for a given image and corresponding will define segmentation. 2.1 Level Set Function The level set, a continuous deformable model method with implicit representation, was developed by Osher and Sethian [5].It is a numerical technique for tracking interfaces and shapes of an object. It is very easy to track shapes with topological changes by the level set method. The level set methods have become very popular in the field of image segmentation due to their ability to represent boundaries of the object that changes with time or are ill defined. At any given time, the level set function simultaneously defines an edge contour and a segmentation of the image.
3. INITIALIZATION OF MASK
Figure 1 A curve C propagating in normal direction [4]. The segmentation is given by inside regions and outside region as shown in figure 1. The minimization problem requires minimizing overall set boundaries C. The level set function is normally a smooth well behaved function. Instead of tracking a curve through time, the level set method evolves a curve by updating the level set function Volume 2, Issue 2 March April 2013 An outstanding characteristic of level set methods is that contours can split or merge as the topology of the level set function changes. Therefore level set methods can detect more than one boundary simultaneously and we can also initialize the multiple contours as a grid . 3.1 Initialization as a Single Mask
Page 286
Figure 5 Variation of Time of Execution with variation in number of contours in grid of segmentation on Teeth2 image. Table 1: Comparative analysis of Single Mask and Grid Mask Initialization for different images with the values of 1=1.0, 2=1.0, =255*255*0.2, =0
Figure 4 (a) Original image (b-f) Segmentation using 6, 9, 12, 15 and 20 contours respectively with 1=1.0, 2=1.0, =255*255*0.2, =0. For this image we get the best segmentation result in 9 contours in a grid .In this case we have kept the number of iteration fixed i.e. 4000.
6. CONCLUSION
We have analyzed the evolution of contour using single mask and grid mask initializations to assess the best approach and performed the Chan-Vese [4] segmentation on real images. The segmented images in both cases are compared on the basis of the convergence time. The evolution of contour is much faster in grid mask. It converges in a few iterations as compared to single mask initialization and gives the acceptable results which are approximate 4 times faster than the single mask initialization.. We have also studied the minimum number of contours in a grid mask that can give the best and acceptable segmentation results. For a particular image, there is some minimum number of contours i.e. optimum number of contours in a grid which gives the Volume 2, Issue 2 March April 2013
ACKNOWLEDGEMENT The author is thankful to Dr. ChanderMohan, Professor in Computer Science and Engineering Department at Ambala College of Engineering and Applied Research, Devsthali, Ambala for their valuable comments and suggestions. References [1] Rafael C. Gonzalez, Richard E. Woods, Digital Image Processing, 2nd ed., Beijing: Publishing House of Electronics Industry, 2007. [2] M. Kass, A. Witkin, and D. Terzopoulos, Snakes: active contour models, Intl J. Comp. Vis., vol. 1, no. 4, pp. 321331, 1987. [3] J. Sethian,Level Set Methods and Fast Marching Methods, Cambridge: Cambridge University Press, 1999. [4] Tony F. Chan and Luminita A Vese, Active Contours Without Edges, IEEE Transactions on Image Processing, Vol. 10 No.2, Feb 2001. Page 288
Page 289