Notes on Pattern counting

• It is usually necessary to clean up the image before counting grains, for example to remove dark areas around the edge and other artifacts in the image, and to maximize the contrast before starting. You should also back up the image (Ctrl-B) before starting.
• Both the threshold and mismatch weight should be increased when counting patterns, compared to counting grains. For example, if the threshold is 0.9, a pixel will match if the average of the red, green, and blue values in a pixel deviate from the pattern by less than (1-0.9)$\times$ 255, or 25.5. If the sum of the signals from each pixel is higher than a minimum signal (defined as pattern area $\times$ threshold), the neuron will register a match.
• Increased speed can be obtained by converting color images to grayscale before starting. Note that just because an image appears to contain only shades of gray, this does not necessarily mean it is a grayscale image.
• Image contrast should be maximized before starting (``Color..Contrast.. Maximize Value''). Increasing the image contrast has an effect similar to increasing the threshold.
• Clicking on ``Enhance grains'' often improves the accuracy of pattern counting as well as the accuracy of grain counting.
• The pattern detection algorithm is currently unable to recognize patterns that are rotated or of different sizes. Often, patterns of different sizes can be recognized by selecting only one edge as the model pattern, e.g. if the desired pattern is a symmetrical round object, drawing the box around the left half of the object instead of the entire object will cause the algorithm to recognize objects of a broad range of sizes.
• The Size Graph is a histogram of the number of objects found that were counted because they were larger than the minimum size, plotted for each possible size (i.e., area of each object in pixels). The Density Graph (or ``Signal Distribution Graph') is a histogram of the number of objects for each possible total signal. The density for each pixel is a number between 0 and 1.0. The total signal is the sum of the densities for all the pixels in a given object. For example, if all the pixels in a grain had a density of 0.5, and the grain was a 10$\times$10 pixel square, the total signal would be 50. This could be different if you calibrated the pixel values (see ``Image Calibration''). The data shown in both of these graphs is compressed to fit onto the screen. To see the complete data set, click on ``Save grain results'' to save the data into a file.