The training phase allows showing a sample image of the item to be matched. For convenience, the sample is usually taken from a rectangular region of interest. Anyway, any pixel can be assigned a "don't care" attribute, meaning that the shape and contents of the pattern can be arbitrary. The pattern center can be freely specified, too.

The search takes place in a rectangular window. An arbitrary number of matching occurrences can be requested. After the search, the positions and scores can be queried.

The pattern matcher is exploiting all the gray-level information available. This means that interpolation techniques allow measuring positions with sub-pixel accuracy, possibly lowering the resolution requirement of the camera sensor.

The pattern matching algorithm uses the well-known normalized correlation technique. Changes in contrast or lightness will not influence the search at all, relieving from the need of strict illumination control.

When the pattern can be confused with other features in the scene, false matches can be discriminated by means of the search score.

For time performance optimization, a few relevant technical parameters are exposed to allow fine tuning and provide better control on the coarse-to-fine searching process.