International Journal of Advancements in Research & Technology, Volume 2, Issue4, April-2013 ISSN 2278-7763 254 CRACK DETECTION IN ARMOURED FIGHTING VEHICLES USING CONTOURLET TRANSFORM ANALYSIS S.Sumathi1 K.S.Lakshmi Narayanan2 ABSTRACT This project aims to develop a methodology to detect cracks with their dimensions in the structure of Armoured Fighting Vehicles using Contourlet Transform analysis. The image analysis based detection is found to be a novel, easy and an inexpensive technique. This project aims in identifying the cracks present in structure of AFVs i.e. Chassis, Turret and sub assemblies like Hydro gas suspension systems etc. The main source of this project includes the images taken using a high resolution CCD camera. Cracks are high frequencies that can be easily detected using contourlet transform which sub bands the image into images of different frequencies using a Laplacian Pyramid (LP) and thereafter decompose the redundancy using Directional Filter Bank (DFB). The extracted features such as mean, standard deviation, variance, mean square error & PSNR help in differentiating the images with good and bad. The project also adds up the technique to measure the crack length and the depth. INTRODUCTION Armoured Fighting vehicle is a combat tracked or wheeled vehicle protected by strong armour having defensive capabilities. In this AFV ,cracks usually occur due to improper welding and casting such as improper penetration,undercutting,porosity,hot Copyright © 2013 SciResPub. U.Nandhini3 J.R.Ramyashri4 tearing, longitudinal cracking, shrinking etc., METHODS OF CRACK DETECTION 1) DYE PENETRANT TESTING - Dye penetrant testing examines the surface of an item for cracks. A liquid penetrant is applied to the surface and left to soak. The liquid is drawn into any cracks via capillary action. The liquid is typically brightly colored or a fluorescent dye. After the soak time has expired, the excess penetrant is wiped off and a developer is applied. The developer is usually a dry white powder suspension that is spayed on the component. The developer is drawn out of the crack by reverse capillary action, resulting in a colored indication on the surface that is broader than the actual flaw, and therefore, much more visible. This technique can be used to detect surface flaws on essentially any non-porous material. 2) RADIOGRAPHY OR X-RAY EXAMINATION - Radiant energy from an X-ray tube or gamma-ray is passed through the section of the object and intensity of emergent rays recorded on a film held on the opposite surface. Defects in the form of voids or cracks are recorded as blackened areas on the film. 3) ULTRASONIC TESTING - It is based on the principle of reflection of high frequency sound waves. If the object contains some defect, the wave is reflected from the surface of the defect and returns in a shorter period of time. If the section is free from defect, the wave is reflected back International Journal of Advancements in Research & Technology, Volume 2, Issue4, April-2013 ISSN 2278-7763 after travelling through the entire section. An oscillograph is used to detect the lengths of time. PROPOSED TECHNIQUE In this project we propose a novel, inexpensive technique for the identification of cracks in Armoured Fighting Vehicle by acquiring the image of the cracked vehicle using high resolution camera. The acquired image is then processed using image processing techniques and cracks are identified using Contourlet analysis and dimension finding by Hough Transform. IMPLEMENTATION Minute cracks are very difficult to identify. Using Contourlet Transform we can find minute cracks in an image. Images are acquired using CCD camera. Before applying CT the image is denoised using 2D Median Filter. Applying the contourlet transform an image is initially decomposed to non-redundant low pass sub image and redundant band pass sub image by LP and the latter when subjected to DFB yields the contourlet coefficients containing the information about the crack. Before applying threshold, the image is subjected to contrast enhancement by adopting Adaptive Histogram Equalization. Edge is detected using canny operator and Threshold for the image can be determined by subtracting the mean and standard deviation value for the image. We assign a value of ‘0’ to the pixel values below the threshold value featuring the cracks and a value of ‘1’ for the pixel values above the threshold value featuring the background. Thus, the resulting binary image shows the cracks of an image isolated from its background. The resulted image is given as input to the Hough transform for calculating length and width of the crack using linear and circular Hough transform. And the classification of the image is done using Support Vector Machine. Thus using this proposed Copyright © 2013 SciResPub. 255 algorithm cracks with dimension in the image can be efficiently detected. In this project we use the mathematical tool Mat lab for obtaining our goal. The code for detecting cracks and dimensions is written in Matlab using the contourlet tool kit. IMAGE ACQUISION USING CAMERA IMAGE PREPROCESSING CONTOURLET APPLICATION IMAGE POSTPROCESSING HOUGH TRANSFORM APPLICATION CONDITION MONITORING BY SVM Figure (a) Flow diagram I) IMAGE ACQUISITION The acquired image by the camera is given as input to the program we have written. Figure (b) Input image II) IMAGE PREPROCESSING The dimension of the image is checked after the acquisition of the image. If the dimension is found to be greater than two the image is converted from colour to gray scale. The gray converted image is then subjected to 2Dimensional Median filter where the image is denoised. Every acquired image has a inherent Gaussian profile present in it. In order to remove it International Journal of Advancements in Research & Technology, Volume 2, Issue4, April-2013 ISSN 2278-7763 256 we denoise the image. The denoised image is assumed to have a higher PSNR compared to the Original image. The gray converted image is given to the median filter. The median is got by sorting all the values from low to high and taking the median of it. The resultant median filtered image is free of salt and pepper noise. Figure (d) Four level contourlet decomposed image. IV) IMAGE POSTPROCESSING Figure (c) Median filtered image III) CONTOURLET APPLICATION The pre-processed image is given as input to the contourlet. The contourlet is a double filter bank structure consists of Laplacian Pyramid and Directional Filter Bank. It is very efficient in handling smooth contours. The Laplacian Pyramid is used for multiscale decomposition and Directional Filter Bank for producing highly directional contourlet Coefficients.The denoised image is subjected to Laplacian Pyramid which decomposes the image into low pass and band pass image. The contourlet Co-efficient are present in the band pass image hence the redundant band pass image is given to the DFB for extracting Contour Co-efficient. The resultant image from the DFB is then post processed by Adaptive Histogram Equalisation. The histogram of grey levels in a window around each pixel is generated first. The cumulative distribution of GL’s, is used to map the input pixel GL’s to output GL’s. If a pixel has a GL lower than all others in the surrounding window the output is maximally black if it has the median value in its window the output is 50% grey. The resultant image is then global threshold and edge detected using canny edge detector. The features such as Mean, Standard Deviation, Variance, PSNR, and Mean Square Error are then calculated from the edge detected image. Figure (e) Thresholded image Copyright © 2013 SciResPub. International Journal of Advancements in Research & Technology, Volume 2, Issue4, April-2013 ISSN 2278-7763 FEATURES VALUE EXTRACTED MEAN 0.0151 STANDARD 0.0070 DEVIATION VARIANCE 1.0556E-007 PSNR 24 MEAN SQUARE 379.8470 ERROR Table (a) First order features 257 LINEAR HOUGH: Each edge point in the (x, y) plane is converted to line in (m,b) plane. First the Hough space is quantised into uniform intervals of accumulator cells and initially all values are set to zero. The dimension of the accumulator is equal to the number of unknowns. In linear Hough transform the dimension of the accumulator is two and in Circular Hough Transform the dimension is three since three unknowns. For each pixel and its neighbourhood, the algorithm evaluates for the existence of edge at that pixel, if so it will calculate the parameter of the line and look accumulator that the parameter fall into, and increment the value of accumulator. The Peak value in the accumulator represents the presence of line. CIRCULAR HOUGH: At each edge point a circle is drawn with desired radius. At the co-ordinates which belong to the parameter of the circle, increment the accumulator cell. The accumulator will now contain the number of circles passing through every individual co-ordinate. The highest element in the accumulator represents the presence of circle. Figure (f) PSNR Plot V) APPLYING HOUGH TRANSFORM Hough transform is used to detect lines, circles and other parametric curves. The input to the Hough transform should be a edge detected image. Hence the resultant image from the post processing is given to the Hough Transform. Hough transform is point to line transformation. Each edge point is converted into line in the parametric space. The numbers of lines intersecting at the point in the parametric space is entered in the accumulator cell. The output of the Hough transform is in the form of accumulator. The Length of the crack is determined using the linear Hough Transform and depth is calculated using Circular Hough Transform. Copyright © 2013 SciResPub. FEATURES IN PPI Crack length 172 Crack depth 12.8471 Table (b) crack length and depth VI) CONDITION MONITORING BY SVM Support Vector machine is used to classify whether the image is normal or abnormal based on the comparison of the trained and testing data. The Support Vector machine takes a set of input data and predicts, for each input, which of the possible classes forms the output, making it a nonprobabilistic BINARY LINEAR CLASSIFIER. The edge detected image is given as input to the Gray level Cooccurrence matrix(GLCM).GLCM is a tabulation of how often different International Journal of Advancements in Research & Technology, Volume 2, Issue4, April-2013 ISSN 2278-7763 combination of gray level occur in the image. GLCM considers the relation between two neighbouring pixels in one offset, as the second order texture, where the first pixel is called reference and the second one the neighbour pixel. Based on the Comparison the GLCM extracts the second order features such as AutoCorrelation,Contrast,Correlation,Cluster Prominence, Cluster Shade,Dissimilarity,Sum of Squares ,Sum average, Sum entropy, Difference Variance, Difference Entropy, Information measure of correlation 1, Information measure of correlation 2,Inverse Difference Normalized,Energy,Maximum Probability and Homogenity are extracted by this GLCM.Since the image is converted to grey image before processing ,each GLCM extracted features has low intensity and high intensity value. These second order features are given as input to the SVM. Based on the trained data the SVM result says the image as good or bad 258 CONCLUSION Thus the crack in the metal image captured is detected using Contourlet transform and its dimensions are measured using Hough transform and the images with and without cracks are classified using Support Vector Machine. REFRENCES [1] Feature Extraction Based On Contourlet Transform And Its Application To Surface Inspection Of Metals By Yonghao Ai;Ke Xu Opt. Eng. 51(11), 113605 (Nov 15, 2012). Doi:10.1117/1.Oe.51.11.113605 [2] M. N. Do And M. Vetterli, “The Contourlet Transform: An Efficient Directional Multiresolution Image Representation,” Ieee Trans. Image Proc., Vol. 14, No. 12, December 2005. [3] T.S. Huang, G.J. Yang, And G.Y. Tang, "A Fast Two-Dimensional Median Filtering Algorithm", Ieee Transactions On Acoustics, Speech, And Signal Processing, 27(1):13-18, 1979. [4] An Evaluation Of The Effectiveness Of Adaptive Histogram Equalization For Contrast Enhancement John B. Zimmerman, Member, Ieee, Stephen M. Pizer, Edward V. Staab, J. Randolph Perry, William Mccartney, And Bradley C. Brenton, Ieee Transactions On Medical Imaging, Vol. 7. No. 4, December 1988. Figure (g) Result of the program [5] A Practical Guide To Support Vector Classification Chih-Wei Hsu, Chih-Chung Chang, And Chih-Jen Lin National Taiwan University, Taipei 106, Taiwan Http://Www.Csie.Ntu.Edu.Tw/~Cjlin Initial Version: 2003 ; Last Updated: April 15, 2010. Figure (h) Result of the program Copyright © 2013 SciResPub. [6] Haralick, R.M., K. Shanmugan, And I. Dinstein, "Textural Features For Image Classification", Ieee Transactions On International Journal of Advancements in Research & Technology, Volume 2, Issue4, April-2013 ISSN 2278-7763 Systems, Man, And Cybernetics, Vol. Smc-3, 1973, Pp. 610-621 [7] Circle Detection Using Hough Transforms Documentation Coms30121 Image Processing And Computer Vision Jaroslav Borovicka - Pinus@Centrum.Cz, Jb2383@Bris.Ac.Uk Copyright © 2013 SciResPub. 259 International Journal of Advancements in Research & Technology, Volume 2, Issue4, April-2013 ISSN 2278-7763 Copyright © 2013 SciResPub. 260