You are on page 1of 4

An Efcient Watershed Segmentation for MR Images

Chaithra MTech(CNE)Student Srinivas Institute of Technology Valachil,Mangalore,India chaithra580@gmail.com Padmanayana Associate. Prof. Dept. of CSE Srinivas Institute of Technology Valachil,Mangalore,India padmanayanabhat@yahoo.co.in

AbstractImage Segmentation has a vital role in the eld of Medical Science especially in the area of cancer research for detecting tumors. Proper diagnosis and effective treatment is only possible if we have an efcient segmenting technique. Due to the rapid calculation speed, watershed transform is developed rapidly in image segmentation eld in recent years. But it has an inevitable over-segmentation problem in the application. In this paper, a new watershed method for MR image segmentation has been proposed. To achieve smoothing and enhancing the contrast in image preprocessing, the total variation model is used as nonlinear lter. Experiments will be performed on real medical MR brain images and kidney images and the new approach is implemented to solve the over-segmentation problem effectively, and to identify brain tumors and the kidney regions in the respective images clearly. Keywords:MR Images,TV Denoising Model,Watershed Transform

I. I NTRODUCTION Image segmentation is used in various applications like medical imaging, locating objects in satellite images, face recognition, trafc control systems, ngerprint recognition and machine vision etc. Medical imaging includes locating tumors and other pathologies, measuring tissue volumes, etc. Segmentation plays an important role in biomedical image processing. Segmentation is the starting point for other processes such as registration, shape analysis, visualization and quantitative analysis. Segmentation of an image is the division or separation of the image into disjoint regions of similar attribute. In clinical practice, Magnetic Resonance Imaging is used to distinguish pathologic tissue from normal tissue, especially for brain related disorders. Manual segmentation by an expert is time consuming and it is very difcult to do accurate segmentation. Hence automatic segmentation algorithms are preferred in diagnostic process. Watershed transform [3] based on mathematical morphology is a popular segmentation method. The main advantages of the watershed method over other previously developed segmentation methods are [4],the resulting boundaries form closed and connected regions, the boundaries of the resulting regions always correspond to contours which appear in the image as obvious contours of objects. However the main limitation is the over-segmentation in the results of the algorithm. Recently a lot of researchers do many works in this eld. Paul R. Hill

et al. [4] developed a novel marker based solution that basins are ooded from selected sources rather than minima, and it was more efcient for image region characterization for content based retrieval. H.P. Ng et al. [5] proposed a new algorithm by performing threshold method on the gradient magnitude image and performing post-segmentation merging on the initial segmentation map. Pousse A et al. [6] presented a method combining a wavelet ltering method with watershed determination, both leading to hot nodule enhancement, results showed it can detect more than 97%. of modules on phantoms. Cha et al. proposed a watershed algorithm based on multiresolution image, use low resolution to reduce the burden of computer, and a new function to merge regions, experiments showed that the method balances calculated efciency and segmentation accuracy. Recently Partial differential equations (PDEs) have led to an entire new eld in image processing and computer vision , it has been widely applied in image smoothing, edge detection, image segmentation el. The total variation [7] image processing model has been generalized and extended in many ways, and has received a lot of attention in last decade. Because the method uses the partial differential equation of the anisotropic diffusion equation for image denoising, it can preserve edges and solve the restoration of the image details. In the paper, we proposed a new watershed segmentation method for MR image based on TV model. To solve the over-segmentation problem of watershed transformation, we use TV denoising model to process the original image before the gradient image composing, use mathematical morphology reconstruction and a threshold method used to extract marker. The experiments will be performed on real medical MR brain and kidney images. II. TV D ENOISING The total variation model was rst proposed by Rudin and Osher. The TV of function has emerged as a very successful regularizer for image denoising and image restored. The main advantages of the resulting denoising method lie on the fact that it allows the preservation of the edges in the image, and its numerical stability and uniqueness property have been shown to be particularly useful in enhancing the effectiveness of the image.

The original TV (ROF) model [7] was proposed by Rudin, Osher and Fatemi. Let u be the original image, f be the noised image,u0 be the additive noise image. So f (x,y)=u(x,y)+u0 (x,y) Based on total variation of the image noise reduction can be the solution of the following minimization problem: min
u

|u | = min
u

u x

u y

dxdy

(1)

Where is the image domain.u denotes the gradient of image u. u [0, ) Subject to the two constraints as follows: udxdy =

f dxdy

(2) (3)

1 (u f )2 dxdy = 2 , > 0 2

Where is the standard deviation of the noise image u0 .Using Lagrange multiplier method to (2) yields the following TV penalized minimization problem: min

|u |dxdy +

(u f )2 dxdy

(4)

Where is a Lagrange multiplier, and it is a parameter to balance the relationship between the two terms. The rst term is regularization parameter to smooth the image, and the second term acts as functional delity measuring parameter to control the difference between u and f. The Euler-Lagrange equation of (4) is given as follow:
2 Let F (u, u ) = 2 (u f ) + |u | then F u

= (u f )
u . | u|

being the divide lines of the domains of attraction of rain falling over the region An alternative approach is to imagine the landscape being immersed in a lake, with holes pierced in local minima.Basins (also called catchment basins) will ll up with water starting at these local minima, and, at points where water coming from different basins would meet, dams are built. When the water level has reached the highest peak in the landscape, the process is stopped. As a result, the landscape is partitioned into regions or basins separated by dams, called watershed lines or simply watersheds. When simulating this process for image segmentation, two approaches may be used: either one rst nds basins, then watersheds by taking a set complement; or one computes a complete partition of the image into basins, and subsequently nds the watersheds by boundary detection.Watershed transform can be denoted as a labelling of the image, such that all points of a given catchment basin have the same unique label, and a special label, distinct from all the labels of the catchment basins, is assigned to all points of the watersheds. An algorithmic denition of the watershed transform by simulated immersion was given by Vincent and Soille [3] for the binary case). Let f : D be digital grey value image, with hmin and hmax the minimum and maximum value of f . Dene a recursion with the grey level h increasing from hmin to hmax , in which the basins associated with the minima of f are successively expanded. Let Xh . Denote the union of the set of basins computed at level h . A connected component of the threshold set Th+1 at level h +1 can be either a new minimum, or an extension of a basin in Xh : in the latter case one computes the geodesic inuence zone of Xh within Th+1 , resulting in an update Xh+1 . Let M INh denote the union of all regional minima at altitude h . Denition1. (Watershed by immersion) Dene the following recursion: Xhmin = {p D|f (p) = hmin } Xh+1 = M INh+1 IZT h+1 (Xh ), h [hmin , hmax ) (6) (7)

Let F u
F (u,u ) u

u |u |

then .(F u )

and

= F u (F u ) = 0 So the Euler-Lagrange equation of (4) is . u |u | + (u f ) = 0


1 u |u n

(5) The watershedW shed(f ) of f is the complement of Xhmin in D : W shed(f ). IV. P ROPOSED M ETHOD A. Preprocessing MR Image is loaded as the input. Loaded image is converted into gray scale format. MR image may consist of lm artifacts and noise in the gradient image results in over segmentation, we use TV denoising model for nonlinear lter, the new model accomplishes the effect of blurring possible noises while enhancing edges . So this pre-processing stage makes the image ready for further processing. B. Application of Sobel Mask and Gradient Magnitude The gradient magnitude is used often to process a grayscale image prior to using the watershed transform for segmentation which is computed using the linear ltering method.

With boundary condition = 0 Where .() is the divergence operator, and when u where diffusion is weak on edges, when u 0where diffusion is strong in smooth region, and in this way it can eliminate noise. III. WATERSHED SEGMENTATION A. Watershed algorithm The watershed transform is the method of choice for image segmentation in the eld of mathematical morphology. In grey scale mathematical morphology the watershed transform improved by Beucher[8] and L. Vincent [3].The watershed transform can be classied as a region-based segmentation approach. In practice, the watershed is applied to the image gradient and the watershed lines separate regions with desired segmentation result. The intuitive idea underlying this method comes from geography: it is that of a landscape, watersheds

A gradient is a two dimensional vector that points to the direction in which the image intensity grows fastest. The gradient magnitude gives the amount of the difference between pixels in the neighborhood which gives the strength of the edge. For any gray scale image(m,n), at co-ordinates (m,n), the gradient vector magnitude and angle at which maximum rate of change of intensity level occurs at the specied coordinates (m,n) can be computed using the equation (8) and equation (9). g (m, n) = g1 (m, n) + g2 (m, n)
2 2

(8) (9)

(m, n) = tan1

g2 (m, n) g1 (m, n)

Where g1 (m, n) and g2 (m, n) are the gradients in the x and y directions. Magnitude of these gradients is computed using the sobel mask H1 and H2

Fig. 1. Sobel Masks

Fig. 2. Watershed Segmentation Process

C. Watershed Transform The basic principle of watershed segmentation is to transform the gradient of a grey level image in a topographic surface, where the values of f (m, n) are interpreted as heights and each local minima embedded in an image is referred as catchments basins.The watershed transformation can be built up by ooding process on a gray tone image. The basic watershed algorithm is well recognized as an efcient morphological segmentation tool which has been used in a variety of gray scale image processes & video processing applications. However, a major problem with the watershed transformation is that it produces a large number of segmented regions in the image around each local minima embedded in the image. A solution to this problem is to introduce markers and ood the gradient image starting from these markers instead of regional minima. The Marker based Watershed Segmentation method possesses several important properties that makes it highly usable for various kinds of image segmentation problems. In the present method, the internal markers are produced from the gray scale image and then external markers are found by nding pixels that are exactly midway between the internal markers. This is done by computing the watershed transform of the distance transformed image of the internal markers. The gradient image is then modied by imposing regional minima at the location of both the internal and external markers. The next step involves the computation of the watershed transformation of the Marker modied gradient image to produce watershed ridge lines. Finally resulting watershed ridge lines are superimposed on the original image and produce the nal segmentation. The proposed method follows these steps which is shown in Fig2. Step1. Use TV model to lter original image. Step2. Calculate the image gradient. Step3. Find foreground marker for the image. Step4. Compute the distance function for back ground marker. Step5. Watershed transform on the modied gradient image. The effectiveness of the proposed method in MR renal image is illustrated in Fig 3.There is over segmentation problem in traditional approaches which is resolved by the proposed method and the proposed method is also effective in detecting the tumor regions in the MR brain image which is shown in Fig4.

Fig. 3. The comparison of different Watershed transformations (a) original watershed method (b) multi-scale morphological method (c) the method proposed in this article

Fig. 4. Tumor detected as white portion

V. C ONCLUSION MR images are playing increasingly important roles in clinics. In this paper, to solve the over-segmentation problem of watershed original segmentation algorithm, we proposed a new watershed method for MR image segmentation.Since MR image is sensitivity to noise, we use TV model as anisotropic diffusion lter for image smoothing,and to enhance the image contrast,the proposed method uses the extracted-marker and distance function to overcome the over segmentation problem. R EFERENCES
[1] Chun-yan Yu,, A Watershed Method for MR Renography Segmentation, ,IEEE 2012. [2] Pratik P. Singhai, Siddharth A. Ladhake, Brain Tumor Detection Using Marker Based Watershed Segmentation from Digital MR ImagesIJITEE 2013. [3] Vicent L, Solille P, Watershed in digital spaces: An efcient algorithm based immersion simulations. IEEE Trans PAMI, 13(6), 1991,pp538598. [4] Paul R. Hill, C. Nishan Canagarajah, David R. Bull, Image Segmentation Using a Texture Gradient Based Watershed Transform, IEEE Transactions on Image Processing, Vol. 12, No. 12, December 2003.pp1618-1633 [5] W.L. Nowinski1, Medical Image Segmentation Using Watershed Segmentation with Texture-Based Region Merging. 30th Annual International IEEE EMBS Conference Vancouver, British Columbia, Canada, August 20-24, 2008.pp4039-4042. [6] Yung-Chieh Lin ; Yu-Pao Tsai ;Yi-Ping Hung, Zen-chung Shih, Comparison between immersion-based and toboggan-based watershed image segmentation , IEEE Transactions on image processing, vol 15 no 3, 2006,pp632-640. [7] L. Rudin,S. Osher. Total variation based image restoration with free local constraints. Proc. IEEE ICIP, volume I, 1994, pp. 31-35. [8] S Beucher, Watershed, hierarchical segmentation and waterfall algorithm,Mathematical Morphology and Its Applications to Image Processing, Kluwer. Academic Publishers, Netherlands, 1994, pp171 203.

You might also like