Counting patterns with neural network method

1. Select ``Process..Grain/pattern counting''.
2. Click on ``Select pattern''
3. With the mouse, click and drag on the image such that the crawling box circumscribes an ``x''.
4. Set Threshold to 0.5.
5. Set Match weight to 1 and Mismatch weight to -0.1.
6. Use the mouse to select a region to count, or do nothing to count the entire image.
7. Click on ``Count patterns''.
8. The x's will be indicated by a red cross as they are counted, leaving only the y's and z's (right panel). If the threshold is too low, the y's and z's will also be counted.
9. The count is displayed in a message box.

Options:

• Select pattern: defines the pattern to search for.
• Read pattern from disk: Define a pattern based on a disk file. (not currently implemented)
• Save pattern to disk: Save current pattern to a file. (not currently implemented)
• Count patterns: Count the occurrences of current pattern in currently-selected region or image, making no assumptions about the prototype pattern. It is necessary to select a pattern before using this option.
• Count grains: Counts occurrences of current pattern, making the assumption that the pattern is supposed to be a dark grain on a light background. It is necessary to select a pattern before using this option.
• Count std. size grains: Uses a pre-defined ``grain'' pattern to count grains. It is not necessary to select a pattern before using this option.
• Enhance grains: Performs a Sobel filtering to improve contrast for grain counting.
• Filename: file name for pattern if saving the pattern to disk file.
• Threshold: Determines the sensitivity to variations in the candidate patterns. A higher threshold means greater selectivity. Too low a threshold will result in extraneous items being counted as patterns, while too high a threshold will cause some patterns to be missed.
• Match Weight: Adjusts the relative weight given to a pixel which is part of a pattern.
• Mismatch Weight: Adjusts the relative weight given to a pixel which is not part of a pattern. This parameter is the penalty for an mismatched pixel, thus it is usually between 0 and -1.