Filler Segmentation of Sem Paper Images Based on Mathematical Morphology

Annali di Chimica, 97, 2007, by Società Chimica Italiana FILLER SEGMENTATION OF SEM 575 PAPER IMAGES BASED ON MATHEMATICAL MORPHOLOGY M. AIT KBIR(°)1 Rachid BENSLIMANE2, Elisabetta PRINCI3, Silvia VICINI3, Enrico PEDEMONTE3 1 Départ. Génie Informatique FST, B.P. 416 Tanger Morocco LTTI, Ecole Supérieure de Technologie, Université Sidi Mohamed Ben Abdellah, P.B 2626- Fes, Morocco 3 Dipartimento di Chimica e Chimica Industriale, Università di Genova, Via Dodecaneso 31, 16146 Genova, Italy 2 Summary - Recent developments in microscopy and image processing have made digital measurements on high-resolution images of fibrous materials possible. This helps to gain a better understanding of the structure and other properties of the material at micro level. In this paper SEM image segmentation based on mathematical morphology is proposed. In fact, paper models images (Whatman, Murillo, Watercolor, Newsprint paper) selected in the context of the Euro Mediterranean PaperTech Project have different distributions of fibers and fillers, caused by the presence of SiAl and CaCO3 particles. It is a microscopy challenge to make filler particles in the sheet distinguishable from the other components of the paper surface. This objectif is reached here by using switable strutural elements and mathematical morphology operators. INTRODUCTION Filler identification and corresponding measurements require cross sectional paper images, which need sample preparation before acquisition. The originality of this work is to quantify the presence of fillers by analyzing images of the paper surface. The morphological observations of the paper samples have been carried out by Scanning electronic microscopy (SEM) associated with EDS microprobe (Energy Dispersive Spectrometry). SEM images of the paper surface and cross sections have been recorded using the Secondary Electron detector at three different magnifications: 500X, 1000X and 2000X. EDS allows determining the elemental composition of each sample, in order to identify the nature of fillers. (°) Corresponding author; Email: aitkbir@fstt.ac.ma AIT KBIR and coworkers 576 A Scanning electron microscope Stereoscan 440 Leica-Cambridge associated with an EDS microprobe Link-Gun Oxford have been used, after metallization of the specimen with a very thin layer of graphite, to obtain a good conductivity. Paper models under study have the following characteristics: - Whatman paper: absence of filler. Presence of fibers with large diameter and comparable size. - Watercolor: presence of a large amount of fillers (Si-Al based) and fiber with heterogeneous dimensions. - Newsprint paper: presence of a large amount of fillers (Si-Al based and CaCO3) and fiber with heterogeneous dimensions. - Murillo: presence of a large amount of fillers (CaCO3) and fiber with heterogeneous dimensions. Fillers formed by SiAl and CaCO3 particles, spread among fibers which involves a weak detection of all edge fibres. Fillers distribution is different from a model of paper to another. The presence of CaCO3 fillers is characterized by the detection of coarse edge fibers with a large amount; it is also shown by the presence of irregular granularity. Whatman model Watercolor model Newsprint model murillo model FIGURE 1. - Paper samples images Fillers segmentation on paper SEM images 577 We aim to quantify automatically the amount of fillers, in SEM images of paper samples. A simple thresholding fails to distinguish automatically between fillers and fibers. This problem can be solved by morphological image analysis, which is based on the idea that images represent a collection of spatial patterns that can be analysed by the way they interact with some predefined patterns called structuring elements. The purpose of the next section is to provide an overview of morphological concepts useful for solving such real image analysis problem. MORPHOLOGICAL ALGORITHMS Mathematical morphology ([1][2]) is based on set theory. The shapes of objects in a gray scale image are represented by object membership sets. This theory can be extended to gray scale images. Morphological operations can simplify image data, preserving the objects’ essential shape characteristics, and can eliminate irrelevant objects. Mathematical morphology is based on two basic operations: dilation, which fills holes and smoothens the contour lines, and erosion, which removes small objects and disconnects objects connected by a small bridge. Such operations are defined in terms of a structuring, a small window that scans the image and transforms the pixels in function of its window content. The first fundamental morphological operation, known as dilation, is defined by: Where F represents the processed image, x a pixel, and SE the structuring element, i.e. the shape used for the analysis. The following figure gives an example of dilation on a 4×4 image with a 3-pixel-wide cross-shaped structuring element. FIGURE 2. - Example of dilation The approach successively centre SE on each pixel of the image and replace the pixel at the centre with the maximum value of the image in the window defined by the SE. The second fundamental morphological operation, known as erosion, does the same thing except that the central pixel is replaced by the minimum value in the window defined by the SE. Here the main idea for designing morphological algorithms consists in the selection, according to morphological theory, of some basic operators like erosion and dilatation. And to combine these operators to build complex algorithms: AIT KBIR and coworkers 578 • internal gradient, external gradient, Beucher gradient, multi-scale gradient; • opening, closing, white top-hat, black top-hat; • hit-or-miss transformation, hit-or-miss opening, hit-ormiss closing, thinning, thickening; • geodesic dilation, geodesic erosion; • reconstruction by dilation, reconstruction by erosion, opening by reconstruction, closing by reconstruction, white top-hat by reconstruction, black top-hat by reconstruction; • regional extrema, h-extrema, h-convex, h-concave, extended extrema; etc. Mathematical morphology finds many applications in the domain of image analysis and segmentation [3] and pattern recognition. Particular attention is presently given to real-time image and video processing, D. Baumann [4] address this problem through hardware implementation on FPGAs. MATHEMATICAL MORPHOLOGY FOR FILLERS DETECTION Filler particles, local peaks, can be processed here by mathematical morphology by using suitable structural elements. Filler Fiber FIGURE 3. - Micrographs of a surface base paper structure Here the opening operation is applied to a profile signal, pixels belonging to a line of the image, to illustrate the possibility of peaks elimination. As we can see gray scale opening fills the hole and eliminates peaks, with size small then the structural element. The tophat transform is a composite operation that uses the morphological opening (for white tophat) to give local maxima in gray scale images that can be used as seeds in a segmentation algorithm. The opening is the operation that removes those bright regions that are smaller in dimension than the SE used in the opening. Then, subtracting this image with the thin peaks cut off from the original image gives you just those peaks, plus some low amplitude noise. The tophat is typically followed by a thresholding operation to have a binary image. To identify fillers, we applied the tophat, using a SE, in 8-connected neighborhood, which is larger than the extent of fillers. Fillers segmentation on paper SEM images 579 FIGURE 4. - Gray scale opening * * * * * * x * * * * * * FIGURE 5. - 12 connectivity structural element Using this structural element, fiber borders, with small size width, are accidentally detected. In fact, these borders are flat in a direction and represent peaks in other directions. This is no the case of fillers which represent peaks in all directions. So, we propose here a new implementation of the white tophat algorithm that uses four directional structural elements: * * x * * * * * x * * * * * x * * x * * FIGURE 6. - four directional structural elements To perform the tophat algorithm, the gray scale opening is slight different here. The basic idea is to perform the gray scale opening, for each pixel, using the four structural elements and affect the 580 AIT KBIR and coworkers processed pixel with the maximum value. This eliminates only peaks related to fillers, which represents approximately the same value of the gray scale opening in all directions. SIMULATION AND RESULTS The two examples bellow show results obtained by using a single structural element that uses 12 neighbourhood pixels. (a) (b) FIGURE 7. - Result of white tophat with 12 connectivity SE and 60 as a threshold. a) whatman sample image b) murillo sample image As we can see here fibers are still present in the segmented image. The following results are related to the second algorithm that uses four different structural elements. Fillers segmentation on paper SEM images 581 (a) (b) FIGURE 8. - Result of white tophat with 4 structural elements and 35 as a threshold. a) Newsprint sample image b) murillo sample image After fillers segmentation, we can determine some their properties such as filler size and distribution. CONCLUSION In order to perform digital measurements on the SEM images of paper each point in the image must be recognized as filler and non filler. For some measures it is necessary to partition the material further, e.g. into individual fibers. The separation of fillers from the rest of the image, commonly called segmentation, is performed here by using mathematical morphology. Depending on fillers size we can find a suitable structural element that can give good detection. Results obtained are very useful to perform digital measurements on high-resolution SEM images. Acknowledgements - This research has been financially supported by the European Community in the framework of the Papertech Project (Project Number: INCO-CT-2004-509095). 582 AIT KBIR and coworkers BIBLIOGRAPHY 1) R.M. Haralick, L.G. Shapiro, Computer and Robot Vision, vol. 1, chapter 5, Addison-Wesley, 1992. 2) A. Ledda and W. Philips "Quantitative Image Analysis with Mathematical Morphology" Third FTW PhD Symposium, Ghent University, Ghent, Belgium: 013, 2002 3) Wayne, Lin Wei-Cheng, “Mathematical Morphology and Its Applications on Image Segmentation”, Dept. of Computer Science and Information Engineering, National Taiwan University, June 7, 2000. 4) D. Baumann, J. Tinembart, "Mathematical Morphology Image Analysis on FPGA" AISTA 2004, Luxembourg, Septembre 2004.