Application of Digital Image Processing Techniques for Asphalt Concrete Mixture Images

Transcription

1 The 12 th International Conference of International Association for Computer Methods and Advances in Geomechanics (IACMAG) 1-6 October, 2008 Goa, India Application of Digital Image Processing Techniques for Asphalt Concrete Mixture Images H.M. Zelelew, PhD Dept. of Civil and Environmental Engineering, Washington State University, Pullman, WA, USA A.T. Papagiannakis, PhD PE Dept. of Civil and Environmental Engineering, University of Texas at San Antonio, TX, USA E. Masad, PhD PE Dept. of Civil Engineering and Texas Transportation Institute, Texas A&M University, College Station, TX, USA Keywords: asphalt concrete, digital image processing ABSTRACT: This paper presents an automated digital image processing (DIP) algorithm called Volumetricsbased Global Minima (VGM) thresholding algorithm for processing asphalt concrete (AC) X-ray computed tomography (CT) images. It utilizes known volumetric properties of AC mixtures as the main criterion for establishing the air-mastic and mastic-aggregate gray scale boundary thresholds. Several DIP techniques were utilized to characterize the AC microstructure. It is well demonstrated that the VGM processed images are significantly improved compared to the raw X-ray CT images. It can be concluded that VGM thresholding algorithm was shown to be a major improvement over the largely manual/subjective techniques used in the past. 1 Introduction Asphalt Concrete (AC) mixtures are uniquely complex heterogeneous materials composed of air voids, mastics and aggregates. Mastics are blends of asphalt binder and fines, typically considered as particles passing sieve No. 200 (i.e., sizes finer than 75 microns). The proportions, distribution, and interactions of these three phases define the micromechanical behavior of AC pavements. The highly complex air void-mastic-aggregate interactions affect significantly their performances and hence their microstructure plays a significant role in modeling and simulation of different behavior AC mixtures. Therefore, realistic representation of the AC microstructure is needed in order to characterize different properties of AC mixtures. The development of high resolution X-ray computed tomography (CT) has demonstrated considerable promise to efficiently capture and characterize the AC microstructure. X-ray CT is a non-destructive advanced imaging technique that generates two- and three-dimensional high resolution images with the capability of capturing the details of the microstructure. Several studies have demonstrated the potential application of X-ray CT technology to characterize different properties of AC mixtures. Recently, it is used to effectively quantify air void distribution, aggregate orientation, segregation and surface texture e.g., Masad et al., (1998); Braz et al., (1999); Masad et al., (1999a, 1999b); Sashidhar (1999); Masad et al., (2001); Tashman et al., (2002); Masad & Button (2004); Wang et al., (2004a, 2004b); and Zelelew & Papagiannakis (2007a). Digital Image Processing (DIP) techniques include image contrast enhancement, image noise removal, thresholding, edge detection and image segmentation. A large volume of literatures deals with characterizing AC X-ray CT images using DIP techniques include Synolakis et al., (1996); Kuo et al., (1998); Persson (1998); Masad et al., (1999a,1999b); Shashidhar (1999); Masad & Button (2000); Masad (2001); Masad et al., (2001); Tashman et al., (2001); Al-Omari et al., (2002); Papagiannakis et al., (2002); Saadeh et al., (2002); Tashman et al., (2002); Banta et al., (2003); Kim et al., (2003); Al-Omari & Masad (2004); and Wang et al., (2004a, 2004b). The majority of the studies highlighted above use a combination of DIP and manual/subjective techniques for processing AC images in a format suited to numerical simulation. Identifying the three phases in AC X-ray CT images has been treated subjectively. Typically, the gray level threshold that separates aggregates from mastics and mastics from air void referred to as thresholding is selected subjectively. These procedures and practices lead to over- or under-estimate the AC mixture constituents that yield unrealistic representation of AC mixture model. Therefore, there is a pressing need to develop an approach that can automatically process the AC X-ray 119

2 CT images using DIP techniques. Consequently, their microstructure is suitable to model/simulate the mechanical behaviour of ACs using numerical simulation techniques such as continuum-based finite element method (FEM) and micromechanical-based discrete element method (DEM). 2 Experimental Data The data analyzed consisted of an AC mixture and its X-ray CT images is part of a Texas DOT funded study (Alvarado et al., 2007). The sample was prepared by the University of Texas-El Paso. The AC core was produced using a typical AC mixture type commonly used in TxDOT districts called Coarse Matrix High Binder Type C (CMHB-C). The aggregate source used was hard limestone (HL) and only one binder grade PG was utilized to prepare the mixture. The gyratory compacted specimen (150 mm diameter by 165 mm height) was cored and sawn to a diameter of 100 mm and a height of 150 mm. The X-ray CT scanning took place at Texas A&M University. The procedures used for capturing the AC images using X-Ray CT are well documented Masad et al., (2002). The AC core was scanned perpendicular to its vertical axis at 1mm distance interval to yield 148 slices per core, ignoring the top and bottom slices. 3 Digital Image Processing (DIP) Zelelew and Papagiannakis (2007a) developed an automated digital image processing (DIP) algorithm called Volumetrics-based Global Minima (VGM) thresholding algorithm for processing the AC X-ray CT images. In this paper, the two most crucial DIP techniques, namely image preprocessing and gray scale thresholding were considered to characterize the AC X-ray CT image microstructure. Readers are referred to Zelelew and Papagiannakis (2007a) for brief descriptions of different types of DIP techniques applicable to characterization and modelling/simulation of AC mixtures that include three-dimensional representation/sectioning, edge detection and segmentation of mastic and aggregate objects. 3.1 Image Preprocessing The two typical image preprocessing procedures include contrast enhancement and noise removal. AC X-ray CT Images consist of pixel representations that vary in gray scale level between 0 (i.e., black object) and 255 (i.e., white objects). Figure 1a shows an example of a raw image consisting of 512 x 512 pixels. The resulting resolution of this image is 195 micron per pixel, which does not allow detecting fine aggregate particles roughly passing sieve No. 70. It is evident that the contrast level of such raw images is very poor. Thus, the most commonly used method called histogram equalization (or linearization) was utilized to increase the quality of the raw image. It consists of adjusting the gray level intensity of pixels to produce a more even distribution throughout the image. (a) (b) (c) Figure 1. Example of Preprocessing an Image of AC Core; (a) Raw Image; (b) Contrast Enhanced Image; (c) Denoised Image; Mixture HL CMHB-C (Zelelew and Papagiannakis, 2007a). Figure 1b shows the enhanced image using histogram equalization. AC X-rayed CT images include a variety of types of noise. Its main sources are sensor quality, as well as image digitizing and preprocessing. Variations in densities within the individual mastic and aggregate also contribute to image noise. Reducing image noise is essential in obtaining enhanced image quality. Median filtering is the most commonly used method to reduce or remove image noises. In median filtering, the gray level of each pixel is replaced by the median of the gray level of all pixel values in the pixel s neighborhood Gonzalez and Woods (2002) and Russ (2002). The local neighborhood is defined by a window of pixels in size, referred to as kernel, where N has typically values of 3, 5, 7, 9 and so on. Several studies implemented the median filtering technique to de-noise the AC X-ray CT image noises Yue et al., (1995); Masad et al., (2001); Yue et al., (2003); and Chandan et al., (2004). In this paper, the median filtering technique was also utilized. A comparison was made for several kernel sizes ranging from 3x3 to 9x9. Better results were obtained using a 3x3 kernel. Figure 1c shows the same AC image discussed previously after de-noising is effected using a 3x3 kernel. The improvement in clarity and contrast between Figure 1a and 1c is significant. 120

3 3.2 Thresholding In this paper, image threshold denotes the gray scale level boundary that separates air voids from mastics and mastics from aggregates. Thresholding is the main thresholding routine accepting as input the preprocessed images and volumetric information of AC mixtures. The method consists of volumetrics-driven thresholding based on a global minima percent error approach that utilizes as thresholding criterion the actual volumetric properties of the AC. It applies to images of cross-sections of AC cores taken perpendicular to the vertical axis at regular distance intervals, in this case 1mm. The algorithm seeks to establish two gray level thresholds, a lower threshold T 1 corresponding to the air void-mastic boundary threshold and a higher threshold T 2 corresponding to the mastic-aggregate boundary threshold. Obviously, pixels with gray level intensities ranging between T 1 and T 2 identify the mastic. For mixture type HL CMHB-C, the corresponding gray scale boundary thresholds were 107 and 158, respectively Zelelew and Papagiannakis (2007a). The absolute errors between the experimentally measured and the VGM estimated volumetric properties of the AC mixture type HL CMHB-C are depicted in Table 1. The maximum absolute error observed was 0.5 % that corresponds to the mastics. This is due to the apparent image resolution (i.e., 195 mm) that does not allow to detecting fine aggregate particles passing sieve No. 70. In general, it can be concluded that the VGM thresholding algorithm is quite accurate in preserving AC mixture volumetric properties and can be implemented to effectively characterize the AC microstructure. Table 1. Comparison of Measured and VGM Estimated Mixture Proportions; Mixture HL CMHB-C. Mixture Components Measured Predicted Absolute Error (%) Air Void (%) Mastic (%) Aggregate (%) HL CMHB-C: Hard Limestone Course Matrix High Binder Type C Mixture Image analysis techniques have been used by researchers to characterize distribution of air voids and assess the gyratory compaction efforts in AC mixtures based on X-ray CT images Masad et al., (2002) and Tashman et al., (2002). Thus, the AC mixture component (i.e., air voids, mastics and aggregates) distribution was characterized utilizing the computed VGM air void-mastic and mastic-aggregate gray scale boundary thresholds (i.e., T 1 =107 and T 2 =158). Figure 2 shows the distribution of air voids, mastics and aggregates for AC mixture type HL CMHB- C. It can be seen from these figures that high percentage of air voids is apparent at the top and bottom regions of the AC mixture core. Relatively, the mastic components distributed uniformly along the height of the mixture. On the other hand, the mixture retains non-uniform distribution of aggregate particles at top and bottom regions of the mixture. This observation is consistent with the findings from the analysis of the directional aggregate segregations by Zelelew and Papagiannakis (2007b) The results of applying the VGM thresholding algorithm to Figure 1 are shown in Figure 3 which highlights each of the three AC phases. This microstructure is especially suited to model or simulate the low temperature cracking behavior of AC mixtures using indirect tension tests (IDT). It should be noted that the VGM thresholding algorithm is robust and versatile and can be adapted in thresholding other composite material X-ray CT images. 121

4 Mixture Proportion (%) Sample Height (mm) Air Void Mastic Aggregate Figure 2. Distribution of Air Voids, Mastics, and Aggregates; Mixture HL CMHB-C. (a) (b) (c) Figure 3. Representation of AC Core Sections (a) Contrast Enhanced and De-Noised Image (b) Air Void Phase in White, (c) Mastic Phase in White, and (d) Aggregate Phase in White; Mixture HL CMHB-C. (d) 4 Conclusions This paper presented the results of the Volumetrics-based Global Minima (VGM) thresholding algorithm, developed by Zelelew and Papagiannakis 2007, implemented on AC X-ray CT images. Several DIP techniques that include image preprocessing and thresholding were utilized to characterize the AC microstructure. The former was used to improve the quality of the low contrasted and noised X-ray CT images whereas, the later was utilized to characterize the air-mastic and mastic-aggregate gray scale boundary thresholds utilizing the volumetric properties of the AC mixtures. The AC mixture proportions and their distributions along the sample height were also characterized. In general, it is well demonstrated that the VGM processed images are significantly improved compared to the raw X-ray CT images and are suitable for inputting their microstructure into numerical simulation techniques such as continuum-based finite element method (FEM) and micromechanical-based discrete element method (DEM). It can be concluded that VGM thresholding algorithm was shown to be a major improvement over the largely manual/subjective techniques used in the past. 122

5 5 Acknowledgements Sincere thanks are expressed to Dr. Eyad Masad of Texas A&M University and to Dr. Soheil Nazarian of the University of Texas-El Paso for supplying the asphalt concrete X-ray CT images and the AC volumetric information, respectively. 6 References Al-Omari A., Masad E Three dimensional simulation of fluid flow in x-ray CT images of porous media. International Journal for Numerical and Analytical Methods in Geomechanics, 28, pp Alvarado C., Mahmoud E., Abdallah I., Masad E., Nazarian S., Langford R., Tandon V., Button J Feasibility of quantifying the role of coarse aggregate strength on resistance to load in HMA. TxDOT Project No and Research Report No Banta L., Cheng K., Zaniewski J Estimation of limestone particle mass from 2D images. Powder Technology, 132, Chandan C., Sivakumar K., Masad E., Fletcher T Application of imaging techniques to geometery analysis of aggregate particles. Journal of Computing in Civil Engineering, 18 (1), Gonzalez R.C., Woods R.E Digital image processing. Upper Saddle River, Prentice-Hall, NJ (USA). Kim D.Y., Park J.W Connectivity-based local adaptive thresholding for carotid artery segmentation using MRA images. Image and Vision Computing, 23, Kim J.B., Kim H.J Multiresolution-based watersheds for efficient image segmentation. Pattern Recognition Letters, 24 (1-3), Kim H., Hass C.T., Rauch A.F D image segmentation of aggregates from laser profiling. Computer-Aided Civil and Infrastructure Engineering, 18, Kim H., Rauch A.F., Hass C.T., Browne C A prototype laser scanner for characterizing size and shape parameters in aggregates. Proceedings of the 9the Annual Symposium, ICAR, Austin, TX. Kuo C-Y., Freeman R.B. 1998a. Image analysis evaluation of aggregates for asphalt concrete mixtures. Transportation Research Record 1615, Transportation Research Board, National Research Council, Washington, D.C., Kuo C-Y., Rollings R.S., Lynch L.N. 1998b. Morphological study of coarse aggregates using image analysis. Journal of Materials in Civil Engineering, 10(3), Masad E., Button J.W Unified imaging approach for measuring aggregate angularity and texture. In: The International Journal of Computer-Aided Civil and Infrastructure Engineering-Advanced Computer Technologies in Transportation Engineering, 15(4), Masad E Review of imaging techniques for characterizing the shape of aggregates used in asphalt mixes. Proceeding 9th Annual International Center for Aggregate Research (ICAR) Symposium (CD-ROM), Austin, Texas. Masad E., Muhunthan B., Shashidhar N., Harman T. 1999a. Internal structure characterization of asphalt concrete using image analysis. Journal of Computing in Civil Engineering, 13(2), Masad E., Muhunthan B., Shashidhar N., Harman, T Aggregate orientation and segregation in asphalt concrete. ASCE Geotech. Special Pub., 85, Masad E., Muhunthan B., Shashidhar N., Harman T. 1999b. Quantifying laboratory compaction effects on the internal structure of asphalt concrete. Transportation Research Record 1681, Transportation Research Board, National Research Council, Washington, D.C., Masad E., Somadevan N., Bahia H.U., Kose S Modeling and experimental measurements of strain distribution in asphalt mixes. Journal of Transportation Engineering, 127(6), Masad E., Jandhyala V.K., Dasgupta N., Somadevan N., Sashidhar N Characterization of air voids distribution in asphalt mixes using X-ray computed tomography. Journal of Materials in Civil Engineering, 14(2), Masad E., Button J Implications of experimental measurements and analysis of the internal structure of Hot-Mix asphalt. Transportation Research Record 1891, Transportation Research Board, Washington D.C., Misiti M., Misiti Y., Oppenheim G., Poggi J.M MATLAB program. The Math Works, Inc., Natick, Massachusetts. Papagiannakis A.T., Abbas A., Masad E Micromechanical analysis of viscoelastic properties of asphalt concrete. Transportation Research Record 1789, Transportation Research Board, National Research Council, Washington, D.C., Persson A Image analysis of shape and size of fine aggregates. Engineering Geology, 50, Russ J.C The image processing handbook. Boca Raton, CRC, FL. (USA). Saadeh S., Tashman L., Masad E., Mogawer W Spatial and directional distributions of aggregates in asphalt mixes. Journal of Testing and Evaluation, ASTM, 30(6), Shashidhar N X-ray tomography of asphalt concrete. Transportation Research Record 1681, Transportation Research 123

6 Board, National Research Council, Washington, D.C., Synolakis C.E., Zhou Z., Leahy R.M Determination of internal deformation field in asphalt cores using x-ray computer tomography. Transportation Research Record 1526, Transportation Research Board, National Research Council, Washington, D.C., Tashman L., Masad E., Peterson B., Saleh H Internal structure analysis of asphalt mixes to improve the simulation of Superpave gyratory compaction to field conditions. Journal of the Association of Asphalt Paving Technologists, 70, Tashman L., Masad E., Angelo J. D., Bukowski J., Harman T X-ray tomography to characterize air void distribution in Superpave gyratory compacted specimens. International Journal of Pavement Engineering, 3(1), Wang L.B., Paul H.S., Harman T., Angelo J.D. 2004a. Characterization of aggregates and asphalt concrete using x-ray computerized tomography: A state-of-the-art report. Journal of the Association of Asphalt Paving Technologists, 73, Wang L.B., Frost J.D., Lai J.S. 2004b. Three-dimensional digital representation of granular material microstructure from x-ray tomography imaging. Journal of Computing in Civil Engineering, 18(1), Yue Z.Q., Bekking W., Morin I Application of digital image processing to quantitative study of asphalt concrete microstructure. Transportation Research Record 1492, Transportation Research Board, National Research Council, Washington, D.C., Yue Z.Q., Chen S., Tham L.G Finite element modeling of geomaterials using digital image processing. Computers and Geotechnics, 30, Zelelew H.M., Papagiannakis A.T. 2007a. A Volumetrics thresholding algorithm for processing asphalt concrete X-ray CT images. International Journal of Pavement Engineering (In-review). Zelelew H.M., Papagiannakis A.T. 2007b. Wavelet-based characterization of aggregate segregation in asphalt concretes. Journal of Materials in Civil Engineering (In-review). 124