The sample preparation techniques for samples to be analyzed using SEM have been described in detail elsewhere [5,6], but will be briefly reviewed here. To prepare a polished specimen for viewing in the SEM, approximately 25 g of the powder to be imaged is blended with an epoxy resin to form an extremely viscous paste. The resin/powder mixture is pressed into a small cylindrical mold and cured at 60 ºC for 24 h. The cured specimen is then cut to obtain a plane surface for imaging.
Saw marks are removed by grinding with 400 grit followed by 600 grit sandpaper. Final polishing is done on a lap wheel with (6, 3, 1, and 0.25) µm diamond paste for 30 s each. After each polishing, the specimen is cleaned using a clean cloth. After the final polishing step, ethanol is used to remove any residual polishing compound. The final polished specimen is coated with carbon to provide a conductive surface for viewing in the SEM.
Once properly prepared, the specimen is placed in the SEM viewing chamber, and signals are collected for the backscattered electrons and X-rays. Typical accelerating voltage and probe current for the backscattered electron images are 12 kV and 2 nA, respectively. For the X-ray images, the probe current is increased to about 10 nA. For analysis of cement powders, in addition to the backscattered electron signal, X-ray images are collected for Ca, Si, Al, Fe, S, K, and Mg. Because these X-ray images are collected at the same location as the backscattered electron image, this series of images can be combined to determine the mineral phase present at each location (pixel) in the two-dimensional image (typically 512 pixels by 400 pixels in size). Typically, magnifications of 250X or 500X are employed for obtaining the SEM and X-ray images. Examples of composite X-ray images for cements 135 and 136 are shown in Figures 3 and 5, respectively. In these images, three different X-ray signal intensities (Ca, Si, and Al) have been mapped into the three color signal channels (red, green, and blue), respectively.
To process the input SEM/X-ray images and to determine the distribution of phases, a decision tree is traversed for each pixel location in the images. An example of a decision tree for a typical cement powder is shown in Figure 2. In this figure, X* represents a critical threshold greylevel value. Pixels having a greylevel greater than the value of X* are considered to contain the element of interest and those with a greylevel below X* are classified as not containing the element. To determine the values of X* for each element, the corresponding greylevel histogram  for each X-ray image is viewed.
After the segmentation tree is traversed, the segmented image produced will still contain a substantial amount of random noise. To improve the image quality, three "filters" are applied in succession to the processed image. First, all isolated one pixel "solid" pixels are converted to porosity. Second, all isolated one pixel "pores" (totally surrounded by solids) are converted to the majority surrounding solid phase. Finally, a median filter  is applied to replace each solid pixel by the majority solid phase present in the surrounding neighborhood, typically a centered 5 pixel x 5 pixel square. This three-fold process removes the remaining noise present in the segmented image, producing an image ready for quantitative stereological analysis, such as those shown in Figures 4 and 6 for cements 135 and 136, respectively. For both cements 135 and 136, two separate image composites were acquired and processed in this manner. These and images for a variety of other cements are available in a prototype online cement images database, being developed within the NIST Partnership for High Performance Concrete Technology program. The prototype database is available for viewing over the Internet at http://ciks.cbt.nist.gov/phpct/database/images.