An improved normalized cross-correlation algorithm for inspection of printed circuit boards - art. no. 635719

Template matching is one of the best and the most widely used pattern recognition method. Normalized cross-correlation (NCC) is the main matching algorithm for template matching method. For templates with significant gray-level variations, also called features, normalized cross-correlation can be a very simple and effective template matching algorithm, even in cases of noisy data and changing lighting level. In the application of automatic optical inspection of printed circuit board, many electronic components have labels on them and can be used as features for cross-correlation template matching. However, there are quite a few components that have no labels, like some type of capacitors and transistors. They are identified by the color or the shade of gray-level instead. These components pose great difficulties for traditional normalized cross-correlation, which will pick up some random variations as features instead and cause false alarms. People-used to deal with this problem by including part of the background into the template to create some artificial features, or by selecting some alternative special algorithms. Both of these methods are not ideal. Because the first method will make the template less universal and subjects to background variations; while the second method will loose many of the nice properties of cross-correlation algorithm. We propose an improved image cross-correlation algorithm, which can recognize both feature based templates and uniform color or gray-level based templates. Compare with the traditional cross-correlation algorithm, this new algorithm can be more accurate and more universal. Experiments have shown that this new algorithm can detect feature-rich, uniformly colored, and uniformly gray templates effectively.