images crack detection technology based on …...images crack detection technology based on improved...

Images Crack Detection Technology based on Improved K-means Algorithm

Cui Fang 1, Li Zhe 2*, and Yao Li 3

1. Information Section, The First Affiliated Hospital, The Fourth Military Medical University, Xi’an Shanxi, 710032, China

2. Information Section, Xiking Hospital, Xi’an Shanxi, 710199, China 3. Xiking Digital Hospital Center, The Fourth Military Medical University, Xi’an Shanxi, 710032, China

*Corresponding author, Email: [email protected]

Abstract—Crack detection is very important to prevent major accident in civil engineering works, but it is still problematic in implementation. The traditional K-means algorithm only takes pixel values into account, which causes the extraction of pavement crack is not accurate. In order to improve the efficiency and accuracy, a novel algorithm is proposed. It is a combination of the improved K-means algorithm and the region growing algorithm, which designs a novel distance function and increases a weight related to crack distance region. The proposed algorithm can effectively abstract the crack information in non-uniform illumination, and improve the performance. The algorithm firstly utilizes histogram algorithm to find the initial clustering center, and then uses the improved K-means algorithm to extract crack. This algorithm overcomes the drawbacks of center indeterminacy and slow speed. Applying the improved K-means algorithm to extract pavement crack image with non-uniform illumination can solve the problem of crack extraction and enhance the reliability and accuracy of pavement crack detection. The results show that compared with traditional K-means algorithm, our proposed algorithm has remarkable effects and can extract the crack information in condition of non-uniform illumination. Index Terms—K-Means Algorithm; Noise; Crack Image; Non-Uniform Illumination; Region Growing Algorithm; Crack Segmentation

I. INTRODUCTION Recently, the road construction is developing rapidly,

while the road maintenance should be highly improved in china. Road maintenance needs to detect the crack information quickly and accurately and abstract its corresponding parameter so as to establish maintenance methods and strategies. The road with non-uniform illumination, to some extent, alters the original texture structure of image, further increases the difficulty of abstracting crack information. Therefore, recently abstracting the crack data parameters in non-uniform illumination are both emphasis and difficulty in the research of road crack image. Traditional pavement crack information is mainly obtained by artificial vision. This mode needs more manpower and has high risk and low efficiency. Moreover, it also can not reach a more accurate evaluation for crack information in non-uniform

illumination. In order to meet the needs of the road maintenance and management under complicated environment, the high performance hardware systems that apply to pavement cracks images processing have gotten a rapid development, and the algorithm of pavement image extraction also has achieved some results. Most of detection systems are composed of CCD industrial camera and lighting facility. France has reached a higher level and Switzerland also has researched and developed CREHOS crack detection system. With the rapid development of highway transportation and the pressing needs of pavement crack image detection, our country also begins to focus on this area. In condition of non-uniform illumination, the developed systems can gain sound effect to pavement crack extraction, but the light device modules of pavement crack detection system often interfered by outside environment. Additionally, these problems such as the quality of imaging environment of light device itself and the instability of the hardware circuit cause the uneven pavement crack images and shadow. Therefore, it is difficult for these systems to process, detect and identify the crack images that have been obtained. So information extraction for pavement crack under uneven light becomes the important research content of highway traffic science and technology area.

In order to address the tough problem of pavement crack extraction with non-uniform illumination, literature [6] applies the method of automatic even light to overcome the shortcoming of crack extraction in condition of non-uniform illumination. This method gets some effect but changes the original state of the crack and loses part of crack pixel information. The literature [7] puts forward the crack detection algorithm based on edge detection Sobel operator which is a new way to detect pavement crack in condition of non-uniform illumination. This algorithm is obviously useful to remove isolated points, but is difficult to detect the pavement cracks that contain most shadow areas. Literature [8] applies rough set algorithm to crack detection area. Although this algorithm improves the function of the clustering algorithm, they cannot get the best effect of pavement crack detection. Because these algorithms randomly select clustering centers and do not optimize and improve

822 JOURNAL OF MULTIMEDIA, VOL. 9, NO. 6, JUNE 2014

© 2014 ACADEMY PUBLISHERdoi:10.4304/jmm.9.6.822-828

them. In order to solve the big randomness problem of applying clustering algorithm to choosing initial centers when detect the pavement cracks; literature [9] adopts recursive binary scale to segment the images and applies Beamlet line segment algorithm to approach the pavement cracks to reduce the impact caused by the uneven light. Although this algorithm improves the effect of cracks extraction to some extent, it is not so sound to the extraction effect of the pavement cracks which have different textures. And it is hard for this algorithm to differentiate the noise such as oil contamination.

Pavement monitoring and maintenance is an expensive yet essential task in maintaining a safe national transportation infrastructure. Traditional monitoring methods use visual inspection of pavements on a regular basis and often require inspectors to travel to the pavement of concern and determine the deterioration level of the crack. Automation of this process may result in great monetary savings and can lead to more frequent inspection cycles. In order to solve the problems of low accuracy and reliability which pavement crack detection systems identify the crack images with non-uniform illumination, the paper put forward the improved K-means algorithm which adds the pixel location information to the influence factors that differentiate crack areas by designing distance functions. The algorithm makes up for deficiency that traditional K-means algorithm only takes pixel values into account. This paper describes an automatic road crack detection method based on image process. By improving the classic traditional clustering algorithm, this algorithm not only promotes its function but also enhances the accuracy of crack extraction. The improved K-means algorithm in this paper yields good results when processing the pavement crack images.

II. K-MEANS ALGORITHM

A. A Basic Introduction to K-means Algorithm The crack information needs to be abstracted in road

image, so the image is quickly calculated, and then transformed to a binary image. Binary images are a specific kind of gray images, which contains two kinds of pixels: black and white. In other words, the road image is divided into two categories: crack target region and background region. The key link of image processing is how to select threshold so as to get binary image. Firstly, the initial center algorithm is adopted to select two pixel values as the initial cluster center. Based on the histogram of crack image, two appropriate cluster center points are selected in this paper. Each pixel point will be partitioned into two sets, according to the nearest distance. The iteration of the process is continued until the clustering result converged. Then, the most appropriate points are selected as clustering result. Lastly, the K-means algorithm is used to traverse all the pixel in the image. Difference between pixel values is employed to measure the amount of similarity within a set of similar data sample, as defined in the following Equation (1)

K

2

1J= | |

j i

j ii x C

x m

(1)

where, jx is data point in the spatial of road crack image, im is the mean of points in cluster ic .

Based on the histogram of crack image, the most appropriate pixel point is selected as cluster center in this paper. Nevertheless, the center point changes all the time in the traditional K-means algorithm, which causes the effects of the abstracted crack information is very uncertain, even worse. With the increase-ment of traversal times, the center point also will change. In additional, the data is very enormous in crack image, which also lead to higher spatial-complexity and more resource consumption.

B. Advantages and Disadvantages of K-means Algorithm Since K-means algorithm is simple and its effect is

good, it is widely applied to many fields, such as Computer Vision, Information Retrieval, Data Mining and Pattern Recognition, and so on, and it can also effectively solve the crack detection problems in pavement. Therefore, K-means algorithm is popular in clustering analysis and achieves good results. When the clustered data in crack image is more dense and there is much difference between pixels value, the K-means algorithm has good effect. When processing the large database, the algorithm is more efficient and relatively stretchable.

But K-means algorithm has its defect, if the category only is defined by pixel value, the crack image data cannot be efficiently handled in non-uniform illumination. The algorithm does not take into account the location information, only their magnitude of pixel value, which is its disadvantage. The effect of the extracted crack is pretty bad under complex background. Because the initial cluster centers are randomly selected, this will cause the algorithm are unstable, and tends to be trapped into local optima. In additional, it is also sensitive to noise of data-set.

III. IMPROVED K-MEANS ALGORITHM AND REGION GROWING ALGORITHM

A. Basic Introduction The proposed algorithm is a combination of the

improved K-means algorithm and the region growing algorithm, which designs a novel distance function and increases a weight related to crack distance region, then is combined with the region growing algorithm. The propose algorithm can effectively abstract the crack information in non-uniform illumination, and improve the performance. In next section, we will introduce he improved K-means algorithm and the region growing algorithm, respectively.

B. Determining Initial Clustering Center It is a key step to select the initial clustering center in

K-means algorithm, but in most cases, the center point is randomly selected, which causes the results have great

JOURNAL OF MULTIMEDIA, VOL. 9, NO. 6, JUNE 2014 823

© 2014 ACADEMY PUBLISHER

randomness and strong uncertainty in the crack image processing. Nevertheless, it is crucial to select some representative pixels in crack image as the center point. In this paper, the crack clustering centers are selected by observing the histogram distribution of pixel value in pavement crack image, which aims at finding out some strong representative pixel points as cluster centers. In other words, it will not only determine the optimal initial cluster centers, but also improve the processing efficiency of pavement cracks. In additional, it also will avoid these disadvantages that the center of crack image is constantly changing in the process of the selection center-point at any time and the cluster centers of crack have small difference in the selection process of the initial cluster centers.

Algorithm idea is to first use the image histogram so as to determine the pixel value of the initial cluster centers, and then combine the improved K-means algorithm with the region growing algorithm, finally finalize the accurate extraction of pavement cracks in non-uniform illumination. The histogram distribution of the original crack in Figure 3(a) is shown as Figure 1, which can determine the cluster center.

0

100

200

300

400

500

600

700

800

900

Grayscale

0 50 100 150 200 250 Figure 1. The histogram distribution of the original crack in Figure

3(a)

C. Distance Information Since a crack image has very complex background in

non-uniform illumination, it is difficult to extracted accurately pavement-cracks. The pixel distance information is introduced into influence factors so as to determine the crack region, which can extract accurately pavement-cracks by comprehensive judgment. In improved K-means algorithm, the similarity measure between data samples is defined as follows:

K

2 2 2

1J= (| | * ( ) ( ) )

i i

j i i ji x C

x m l i R j R

(2)

where, i,j denote the horizontal coordinate and vertical coordinate of pixel, respectively; l is adaptive coefficient of crack position information, which is adjusted according to different crack image; iR is the average of horizontal coordinates in crack region; jR is the average of vertical coordinates in crack region; jx is data point in the spatial of road crack image, im is the

mean of points in cluster ic , and the center point is selected by histogram in paper.

Sum square error (SSE) is used as criterion function to evaluate the performance in improved K-means algorithm. Given a data-set X, it only contain not categorical attribute, but description attribute. Assume X contains k clusters 1 2 3{ , , ...... }kX X X X , while the crack image only has two clusters in paper; the quantity of samples in each cluster is 1 2 3, , ...... kn n n n , respectively; p is cluster center selected by histogram; the average in each cluster is

1 2 3, , ...... km m m m , respectively. The lower the error E is, the better performance the cluster algorithm has, as shown in Equation (2):

2

1| |

i

k

ii p X

E p m

(3)

C. Region Growing Algorithm 1) Algorithm Principle Region growing is a simple region-based image

segmentation method. It is also classified as a pixel-based image segmentation method since it involves the selection of initial seed points. It mainly examines neighborhood pixels of initial “seed points” and determines whether the pixel neighbors should be added to the region. Therefore, region growing selects seed points based on the preset pixel value in crack to develop large regions crack set. Seed point selection is based on some criterion, such as pixels value, gray-level range, etc. The regions are grown from these seed points to adjacent points depending on the region criterion. The initial region begins as the exact location of these seeds in pavement crack image. Then, a qualified pixel value is merged to the corresponding region. Lastly, it is necessary to go through all the pixels in pavement crack image. It is an iterated process until there is no change in two successive iterative stages.

2) Seed Point Selection Since the crack pixel in pavement image has a zonal

shape and pixel-value of crack is changed quite rapidly in non-uniform illumination, the seed point is selected by traversing the entire pixel in the image. More concretely, pixel-value of crack in each row is first traversed to count the number of pixels less than 15. Then, pixel-value of crack in each column is also traversed to count the number of pixels less than 15. Finally, any a point less than threshold are selected as initial seed point in the row with the maximum number of all of the rows.

3) Algorithm Idea By means of integrating the pixel of pavement crack

with similar feature in non-uniform illumination, the crack region eventually is segmented, which is the basic theory of region growing algorithm. Assume an M×N crack image with non-uniform illumination is used, where ( , )G i j denotes the gray-scale value of coordinates ( , )i j on crack pixel. In additional, given a mark ,i jF , , 0i jF denotes ( , )i j does not belong to the crack region based on region growing algorithm,



while , 1i jF represents the pixel is from the crack region and has been covered. As for the crack pixel ( , 1)i j from 4-connected neighborhood of crack pixel ( , )i j , if the following inequalities are hold

( ( , ), ( , 1)) (0 ,0 1 )S G i j G i j T i M j N (4)

, 1 0i jF (5)

where, ( ( , ), ( , 1))S G i j G i j is to measure the degree of similarity between crack pixel and . So we can get

, 1 1i jF , and ( , 1)i j is become the current seed pixel for searching the next pixel. That is, we keep examining the adjacent pixels of seed points. If they have the same intensity value with the seed points, we classify them into the seed points. The similarity degree is defined as the difference between gray-valueof two pixels in paper, T is a threshold selected as 20 by simulation and the growing criterion is expressed in S, so we can get as the following Equation:

| ( , ) ( , 1) |G i j G i j T (6)

After the pavement with non-uniform illumination is processed by clustering algorithm, the region growing criterion shall be designed to abstract the crack information and it would directly influence the accuracy and efficiency of crack detection in conditions of non-uniform illumination. The growing criterion is designed for evaluating ascription degree between crack region and seed point. In general, the distance between crack pixel and seed point is adopted as the region growing criterion, while we choose the gray value of crack pixel and seed point to estimate their similarity.

E. Morphological Processing By taking above steps, the exact region of crack can be

abstracted, but there are still some burr. In order to reduce the influence brought by noise, the crack region with non-uniform illumination firstly is transformed into binary image. Then we use morphological image processing and spatial filter processing to improve the quality of binary segmentation image. In addition, dilation and erosion are used to erase the burrs and noises in the images. The erosion structure element A and dilation structure element B are first constructed, then the binary image C of pavement crack is traversed. When the center point of structure element is moved to point ( , )x y in binary image, the boundary of crack image can be obtained by first dilation, and then erosion, where the structure element is set as 3*3.

F. Algorithm Flowchart Given a crack image with non-uniform illumination

shown in Figure 3(a), the sample space of crack images are denoted as 1,1 1,2 1,3 1,4 ,{ , , , ...... }i jX X X X X . Figure 2 is the flowchart of our proposed algorithm. The basic flow of image processing is described as follows:

(1) Calculate the histogram of sample space 1,1 1,2 1,3 1,4 ,{ , , , ...... }i jX X X X X in crack image, and select

the peak ,Zi j of histogramas clustering center. (2) Calculate the distance between pixel value and seed

point, then judge whether the point is from crack region based on measure criterion.

(3) Traverse all the pixel in crack image point to select seed point.

(4) Use region growing criterion and region growing algorithm to determine crack region.

(5) Use morphological image processing to achieve the exact region of crack image.

Figure 2. Algorithm flowchart

IV. EXPERIMENTAL RESULTS AND ANALYSIS The crack image is typical and complex in paper. In

additional, the crack image with non-uniform illumination has complex background and is seriously effected by noise. It is very difficult to obtain satisfied result by the traditional threshold segmentation algorithms. The experimental results are shown in the following section.

A. Comparison K-means Algorithm with Improved Algorithm

You can see from Table 1 that the improved algorithm has a marked improvement in time and iterative times. It is because the improved algorithm optimizes the selection of center point and calculation process.

TABLE I. COMPARISON OF ALGORITHM

Algorithm Time(s) iterations K-means algorithm 17.9255 21 Improved algorithm 15.0007 14

In the experiments, all of the used images are crack

images with non-uniform illumination. The experiment



results are showed in the following figure. Figure 3 (a)-(d) is, respectively: (a) pavement images with non-uniform illumination; (b) Result of traditional K-means algorithm; (C) Coarse result of improved algorithm; (d) Result of our proposed algorithm. As we can see from experimental results, our proposed algorithm has remarkable effects and can extract the crack information with non-uniform illumination.

(a) (b)

(c) (d)

Figure 3. Visual comparison

(a) (b)

(c) (d)

(e) (f)

Figure 4. Experimental results of other algorithms

Figure 3(a) is a crack image with non-uniform illumination and is proposed by other algorithm, as shown in Figure 4. Figure 4 (a)-(e) is, respectively: (a) Result of rough set algorithm, which we can see that the shadow region is not removed and the structure of crack image is changed; (b) Result of uniform light algorithm, which loses some of crack information; (C) Result of Sobel operator, which do not exactly extract the crack region; (d) Result of Robert operator, which has significantly changed the crack information. (e) Result of LOG operator, which the shadow region is not removed; (f) Result of Prewitt operator, which the effect is not so obvious;

As we can see from experimental results, compared with other algorithms, our proposed algorithm has remarkable effects and high accuracy, and can extract the crack information with non-uniform illumination, which demonstrates the effectiveness of the proposed method in preserving and recovering image structure such as edges and textures. In additional, the algorithm can also smooth the effect of noise. The obtained result covers most regions of crack image, and substantially remove the illumination region. The detected crack information is accurate and reliable, and can truly reflect the morphological characteristics of cracks, so it establishes the foundation for accurately extracting the crack parameter.

V. CONCLUSION The improved K-means algorithm can accurately

extract the pavement crack information under the non-uniform illumination. The algorithm firstly utilizes histogram algorithm to find the initial clustering center, and then use the improved K-means algorithm to extract crack. This algorithm overcomes the drawbacks of indeterminacy of center identification and slow speed. Applying the improved K-means algorithm to extract pavement crack image with non-uniform illumination can solve the problem of accurate crack extraction under different circumstances and enhance the reliability and accuracy of pavement crack detection. In addition, the algorithm is simple and its processing rate is also fast. It also can decrease the interference of detection result under non-uniform illumination and meet the requirements of the pavement crack image extraction. Therefore, the improved K-means algorithm proposed in this paper is of great significance to the development and application of intelligent pavement crack detection systems in condition of non-uniform illumination.

It can be seen from the segmentation result which the main crack of the image in non-uniform illumination is relatively clear and is not affected by non-uniform illumination. But the improved algorithm just applies to the pavement crack image that has single cracks in single image with non-uniform illumination. In practical application, although the algorithm applies to most of pavement crack images in condition of non-uniform illumination, it is not so precise. So the future research direction is how to solve the extraction problems of number and parameter from multi-image and multi-crack.



REFERENCES

[1] Narad, Charles E., et al. "Packet processing system including a policy engine having a classification unit." U.S. Patent No. 6,157,955. 5 Dec. 2000.

[2] Gupta, Pankaj, and Nick McKeown. "Algorithms for packet classification." Network, IEEE 15.2 (2001): 24-32.

[3] Modersitzki, Jan. Numerical Methods for Image Registration (Numerical Mathematics and Scientific Computation). USA: Oxford university press, 2004.

[4] Zitova, Barbara, and Jan Flusser. "Image registration methods: a survey." Image and vision computing 21.11 (2003): 977-1000.

[5] Reddy, B. Srinivasa, and B. N. Chatterji. "An FFT-based technique for translation, rotation, and scale-invariant image registration." Image Processing, IEEE Transactions on 5.8 (1996): 1266-1271.

[6] Pluim, Josien PW, JB Antoine Maintz, and Max A. Viergever. "Image registration by maximization of combined mutual information and gradient information." Medical Image Computing and Computer-Assisted Intervention–MICCAI 2000. Springer Berlin Heidelberg, 2000.

[7] Pluim, Josien PW, J. B. Antoine Maintz, and Max A. Viergever. "Interpolation artefacts in mutual information-based image registration." Computer vision and image understanding 77.2 (2000): 211-232.

[8] Ashburner, John, et al. "Incorporating prior knowledge into image registration." Neuroimage 6.4 (1997): 344-352.

[9] Cox, Robert W., and Andrzej Jesmanowicz. "Real-time 3D image registration for functional MRI." Magnetic resonance in medicine 42.6 (1999): 1014-1018.

[10] Roche, Alexis, et al. "The correlation ratio as a new similarity measure for multimodal image registration." Medical Image Computing and Computer-Assisted Interventation—MICCAI’98. Springer Berlin Heidelberg, 1998. 1115-1124.

[11] Woods, Roger P., et al. "Automated image registration: II. Intersubject validation of linear and nonlinear models." Journal of computer assisted tomography 22.1 (1998): 153-165.

[12] Mattes, David, et al. "PET-CT image registration in the chest using free-form deformations." Medical Imaging, IEEE Transactions on 22.1 (2003): 120-128.

[13] Thévenaz, Philippe, and Michael Unser. "Optimization of mutual information for multiresolution image registration." Image Processing, IEEE Transactions on 9.12 (2000): 2083-2099.

[14] Yoo, Terry S. Insight into images: principles and practice for segmentation, registration, and image analysis. Vol. 203. Wesllesley^ eMassachusetts Massachusetts: AK Peters, 2004.

[15] Christensen, Gary E., and Hans J. Johnson. "Consistent image registration." Medical Imaging, IEEE Transactions on 20.7 (2001): 568-582.

[16] Gonzalez, Rafael C., Richard E. Woods, and Steven L. Eddins. Digital image processing using MATLAB. Vol. 2. Tennessee: Gatesmark Publishing, 2009.

[17] Guizar-Sicairos, Manuel, Samuel T. Thurman, and James R. Fienup. "Efficient subpixel image registration algorithms." Optics letters 33.2 (2008): 156-158.

[18] Transformation-Alpert, Principal Axes. "The principal axes transformation-a method for image registration." J Nucl Med 31 (1990): 1717-1723.

[19] Rohde, Gustavo K., Akram Aldroubi, and Benoit M. Dawant. "The adaptive bases algorithm for intensity-based

nonrigid image registration." Medical Imaging, IEEE Transactions on 22.11 (2003): 1470-1479.

[20] Dai, Xiaolong, and Siamak Khorram. "A feature-based image registration algorithm using improved chain-code representation combined with invariant moments." Geoscience and Remote Sensing, IEEE Transactions on 37.5 (1999): 2351-2362.

[21] Zhong S, Xia K, Yin X, et al. The representation and simulation for reasoning about action based on Colored Petri Net// Information Management and Engineering (ICIME), 2010 The 2nd IEEE International Conference on. IEEE, 2010: 480-483.

[22] Le, D., Jin, Y., Xia, K., & Bai, G. (2010, March). Adaptive error control mechanism based on link layer frame importance valuation for wireless multimedia sensor networks. In Advanced Computer Control (ICACC), 2010 2nd International Conference on (Vol. 1, pp. 465-470). IEEE.

[23] Xia K, Cai J, Wu Y. Research on Improved Network Data Fault-Tolerant Transmission Optimization Algorithm. Journal of Convergence Information Technology, 2012, 7(19).

[24] XIA, K., WU, Y., REN, X., & JIN, Y. (2013). Research in Clustering Algorithm for Diseases Analysis. Journal of Networks, 8(7), 1632-1639.

[25] Yao, Yufeng, Jinyi Chang, and Kaijian Xia. "A case of parallel eeg data processing upon a beowulf cluster." Parallel and Distributed Systems (ICPADS), 2009 15th International Conference on. IEEE, 2009.

[26] Kai-jian, Xia, et al. "An edge detection improved algorithm based on morphology and wavelet transform." Computer and Automation Engineering (ICCAE), 2010 The 2nd International Conference on. Vol. 1. IEEE, 2010.

[27] Wolberg, George, and Siavash Zokai. "Robust image registration using log-polar transform." Image Processing, 2000. Proceedings. 2000 International Conference on. Vol. 1. IEEE, 2000.

[28] Wang, He, et al. "Validation of an accelerated'demons' algorithm for deformable image registration in radiation therapy." Physics in Medicine and Biology 50.12 (2005): 2887.

[29] Ardekani, Babak A., et al. "A fully automatic multimodality image registration algorithm." Journal of computer assisted tomography 19.4 (1995): 615.

[30] Avants, Brian B., et al. "Symmetric diffeomorphic image registration with cross-correlation: Evaluating automated labeling of elderly and neurodegenerative brain." Medical image analysis 12.1 (2008): 26-41.

[31] Lemieux, L., et al. "A patient-to-computed-tomography image registration method based on digitally reconstructed radiographs." Medical physics 21 (1994): 1749.

[32] Fischer, Bernd, and Jan Modersitzki. "Curvature based image registration." Journal of Mathematical Imaging and Vision 18.1 (2003): 81-85.

[33] Thirion, Jean-Philippe. "New feature points based on geometric invariants for 3D image registration." International journal of computer vision 18.2 (1996): 121-137.

[34] Vercauteren, Tom, et al. "Non-parametric diffeomorphic image registration with the demons algorithm." Medical Image Computing and Computer-Assisted Intervention–MICCAI 2007. Springer Berlin Heidelberg, 2007. 319-326.

[35] Brock, Kristy K., et al. "Feasibility of a novel deformable image registration technique to facilitate classification, targeting, and monitoring of tumor and normal tissue."



International journal of radiation oncology, biology, physics 64.4 (2006): 1245.

[36] Salvi, Joaquim, et al. "A review of recent range image registration methods with accuracy evaluation." Image and Vision Computing 25.5 (2007): 578-596.

[37] Goshtasby, A. Ardeshir. 2-D and 3-D image registration: for medical, remote sensing, and industrial applications. Wiley-Interscience, 2005.

[38] Goshtasby, Ardeshir, George C. Stockman, and Carl V. Page. "A region-based approach to digital image registration with subpixel accuracy." Geoscience and Remote Sensing, IEEE Transactions on 3 (1986): 390-399.

Cui Fang, female, (1976-), come from Jiangsu province China, bachelor’s degree, Intermediate engineer, working at Information section, the first affiliated hospital, the fourth military medical university, major in Biological engineering.

Corresponding author: Li Zhe, male, (1964-), come from Shanxi province China, bachelor’s degree, Associate professor, Deputy director of the physician, working at Information section, Xiking hospital, major in the management of hospital informationization. Yao Li, born in1983,male, be from Xi’an Shanxi China, Master of education, intermediate engineer, work at the Digital Center of Xiking Hospital, the Fourth Military Medical University, research on hospital network management and network security. From 2010~2013, he studied at Xidian University whose major was software engineering for master. From 2011.12. to present, he works at the digital center of Xiking hospital, the fourth military medical university as engineer. In 2012 part of xiking hospital disciplines help project the field medical team based on Internet of things application research of dynamic management of war readiness material ", a work for four counts of software copyright.



images crack detection technology based on …...images crack detection technology based on improved...

Documents