Samples of extruded products have been imaged and processed using image segmentation techniques developed within the Image Analysis Project of the CSIRO Mathematical and Informaton Sciences. If product quality can be characterised by the visual appearance of the product, we anticipate that image analysis techniques can be used to extract the salient visual features. These features could then be analysed to provide a quantitative and objective measure of product quality. We present here the imaging technique used to measure pore size distribution in a sample of an extruded product.
To begin with, the sample is prepared for imaging. As shown in Fig. 1, at present we simply obtain a thin slice of the sample by cutting it with a razor blade.
|Sample cut with razor blade(a)||Sample used for imaging(b)|
Figure 1: Sample preparation (a) The sample is cut with a razor blade (b) A thin slice of the sample is used for imagery
The next step is to obtain an image of the slice of the sample. We have constructed our own simple image-capturing set-up, consisting of lights, a black and white CCD camera and a PC with a frame-grabber board. The lighting used is a circular light which surrounds the sample, thus giving a diffuse light source. The resulting captured image is shown in Fig. 2. In this image one can see that the sample consists of pores surrounded by walls. The pores are usually grey, but as the sample is a thin slice they often manifest as holes in the sample, letting the black background show through. The walls that surround the pores are brighter and we use this information to identify the pores in the sample.
Figure 2: An image of the sample
We then process this image to find the features that we are interested in (this process is called image segmentation ). At present, we have only targeted the pores in the sample. In Fig. 3 is shown the resulting segmentation, where the white objects in this binary image are considered as estimates of cross-sections of pores in the sample. This segmentation was obtained by using morphological image processing operations such as openings and closings.
Once the image is segmented, we are free to derive summary statistics from the segmented features, which in our case are the cross-sections of pores. We can illustrate the kind of summary statistics that can be produced using the example of pore area. In Fig. 4 is an area-weighted histogram illustrating pore area information in the sample. Note that the cross-sectional area is the area of the pore as it appears from above and not the three-dimensional volume of the pore. Along the x-axis of the histogram is the area of pores in square millimetres and along the y-axis is the area fraction of pores in the batch which are of this area. One can see from this figure that cross-sectional areas are mostly less than 1 square millimetre, although some are bigger than 1.5 and the biggest is around 2.5 square millimetres. Pore area in square millimetres can be determined from the pore area in pixels by simply taking an image of a ruler and then deriving the number of pixels per millimetre in this image. Based on the results from a commercial-in-confidence feasibility study, we found that such a histogram is a good discriminator for the quality of the particular product that we analysed during this study.