基于形态学的分水岭算法牙科X射线图像分割.docx

1、基于形态学的分水岭算法牙科X射线图像分割英文原文Watershed Algorithm Based on Morphology for Dental X-Ray Images SegmentationAbstractA new watershed algorithm based on mathematical morphology, which can be applied to dental X-ray images segmentation, is proposed in the paper. In order to separate each tooth and improve the

2、serious problem of over-segmentation of traditional watershed algorithm, we apply the processing based on morphology to get the images enhanced before watershed algorithm. First, using the top-hat-bottom-hat transformation to amplify the contrast of foreground and background and remove noises. And t

3、hen erosion procedure is used to weaken the degree of adhesion between teeth. Hole-filling can help to eliminate some unnecessary split-line which can cause the phenomenon of wrong segmentation. Finally we apply the watershed algorithm using distance transformation of the binary image to get the seg

4、mentation result. From the experiment results, we can see that this algorithm can separate teeth accurately and overcome the over-segmentation efficiently compared with traditional method.Keywords-watershed algorithm; mathematical morphology; dental X-ray image; image segmentation; distance transfor

5、mationI. INTRODUCTIONDental biometrics is an appropriate method to identify deceased individuals of disasters such as criminal cases, blasts and Tsunamis, because teeth with their dental works are very resistant to modest force effects and high temperature and chemical corrosion and also possess goo

6、d biometric properties. And it also provides important information in dental medical treatment for further analysis to get the diagnostic. Image segmentation as a precursor has been seen as one of the most important aspect of the entire process like pattern recognition. It gives clarity information

7、and the levels to subdivision depend on the problem being solved.Image segmentation as an indispensable method is applied to extract quantitative information of specific tissues, also as the premise and a step to realize the processing of visualization.And it is still a challenging problem in comput

8、er vision and image processing. Areas obtained by segmentation are both independent from each other. And all the areas belong to one region will share the special consistency. The objective of segmentation is to localize the region of each tooth in dental X- ray image 3. At present, image segmentati

9、on has been broadly divided into 3 categories: cluster analysis, edge detection and region extraction. Each of those methods has got disadvantages such as the failure to get the number of clusters before segmentation when applying cluster analysis.Watershed algorithm based on morphological theory is

10、 a kind of nonlinear segmentation. It possesses great qualities in positioning the edge rapidly and accurately and behaves excellent at detecting objects with weak edges etc. But traditional watershed algorithm has serious problem of over- segmentation under the influence of noise and the fine textu

11、re of the objects. A variety of approaches on how to overcome the split have been proposed for different purposes in different applications.Watershed Segmentation Algorithm based on morphology has been applied to cellular images, as in 1, aimed at reducing the over-segmentation. It requires fewer co

12、mputation and simpler parameters and gets segmentation results more accurately compared with other tradition methods.In 2 and 3, Nomir and Adbel-Mottaleb introduced a fully automated segmentation technique. It starts by applying iterative thresholding followed by adaptive thresholding to segment the

13、 teeth from both the background and the bone areas. And after adaptive thresholding, horizontal integral projection followed by vertical integral projection are applied to separate each individual tooth. And this method can achieve the position of each tooth precisely.In 4, an algorithm based on wav

14、elet transform (WT) to segment dental X-ray images was proposed. It contains three major steps: dental X-ray preparing, panoramic radiograph segmentation using wavelet transformation and enhancement image with morphological image processing. And the segmentation result using this method is better th

15、an thresholding segmentation and adaptive thresholding segmentation.In 7, a watershed segmentation algorithm based on grayscale morphological pretreating is presented. Opening operations are applied to remove small light details before applying watershed algorithm. Therefore the phenomenon of over-s

16、egmentation was controlled and the touching objects were segmented precisely.In this paper, we introduce a new method based on the performance of watershed algorithm and the characteristics of dental X-ray images. At first, we apply the bottom-hat -top-hat transformation to enhance the dental radiog

17、raphs. Then we used the erosion algorithm to weaken the degree of adhesion between teeth and remove the noises. And the imfill() function could help to eliminate the possibility of over-segmentation caused by the upcoming processing. Finally we utilize the watershed algorithm using distance transfor

18、m of the binary image to get the segmentation result.II. WATERSHED ALGORITHM BASED ON MORPHOLOGYA. Top-hat-bottom-hat transformationTop-hat transformation is defined as the difference of the original image minus the image applied with the operation of imopen to remove the some points with the highes

19、t gray value of the image. And the bottom-hat transform is the difference of the image applied with the operation of imclose to remove the points with smallest gray value and then minus the original image. The definitions are as follows: Top-hat transformation：Bottom-hat transformation:Where f is th

20、e original image and b represents the structure elements.The opening of f by b , denoted f b, isThe closing of f by b , denoted f b, isWhere is the operation of erosion and represents dilation.Opening operation is used to remove regions smaller than the structure elements with relative high gray val

21、ues while the closing operation can remove the small regions with relative low gray values. And these two operations possess great quality in remaining all the levels of gray value and keeping the large regions with high values relatively unchanged. One of the important usages is that these two tran

22、sformations are capable of correcting uneven illumination effects. Top-hat transform can be used in situation of bright objects against a dark background, while the bottom-hat transform is applied in the opposite situation.Top-hat transformation has some certain characteristics of high pass filter t

23、hat it can be used to highlight the gray peak and enhance the edge information of the targets. And the bottom-hat transformation can be applied to get the valley of the gray value and prominent the boundaries between connected targets like teeth. Therefore, these two transformations can be used in c

24、ombination to get the effect of image enhancement for the foreground and background gray are further stretched as well as the objectives and details are highlighted.In this paper we can get the image with its contrast effectively improved by adding the original image with the image applied with top-

25、hat transformation and then minus the image applied with the bottom-hat transformation.After enhancement the gray image will be transformed into a binary image by using thresholding for the next processing.B. The morphological processing to the binary imageMathematical morphology is very useful as a

26、 tool to extract image components applied in the representation and description of the boundary, the bones and the convex hull. Erosion and dilation are the basic operations of morphology. This paper mainly uses the processing of erosion which actually means using structural elements to fill the ima

27、ge. Erosion would shrink or refinement the objections. The mode and degree are controlled by the structural elements. We apply the structural elements of b to fill the collection A . And we can assume that the collection A is eroded by b if A still contains the structure elements b after the filling

28、s. It is defined as follows:Where b represents the structure elements.Difficulty exists in dental segmentation because of the adhesion between teeth. We could not get the position of each tooth precisely without delete the adhesion. Therefore, we need to apply the image with erosion operation to wea

29、ken the degree of adhesion between teeth for the benefit of segmentation.Hole-filling refers to the image contour filling. Usually the inner contour area should not be bigger than the maximum area of targets. There may be some holes which should not exist when the dental X-ray images are transformed

30、 into binary images under the influence of shape characteristics of tooth and the non-uniform gray distribution. And these holes may be transformed into little areas independent of each other after erosion. This would lead to the phenomenon of over-segment after applied watershed algorithm. Therefor

31、e, the structure elements b cannot be too large for the possibility of holes generated. And it also cannot be too small for the adhesion must be deleted completely. In this paper, we apply the operation of erosion for two times and during these two procedures add the operation of imfill() to avoid t

32、he generation of over-segmentation caused by holes.C. Watershed segmentation algorithm using distance transformation of binary imageDistance transformation turns the binary image into a gray image. And the value of location ( x, y )is the distance of pixel to its nearest background pixel. It aims to distinguish the boundary pixels and the inner pixels.Watershed algorithm based on morphological theory is first proposed by S. Beucher and L. Vincent and developed rapidly in image segmentation field in recent years 6- 7. In geography, a watershed refers to a dam. The river systems in t