Well-defined background image content and/or well-defined target characteristics exist in many machine vision applications. Machine vision systems use background images or target templates as reference information for detection, matching, inspection and measurement. Conventional approaches rely on background images or target templates specified by users for references. Unfortunately, this works only in simple, well-controlled applications. It fails in applications with significant variation.

DRVision Technologies LLC reference learning technology improves the repeatability, efficiency, usability and effectiveness of reference information. It improves the outcome of reference-based applications that encounter significant variation.

DRVision Technologies LLC reference learning technology automatically extracts useful data from images to represent reference information and expected application variation. This automatic approach produces data that represents the application better and more consistently than data that is manually selected. During application weights of object regions with low variation are heightened, and the weights of object regions with high variation are lowered, increasing application robustness. If necessary, DRVision Technologies LLC structure-guided processing technology can be used to compensate for the loss in discrimination power identified by the reference learning technology. This approach circumvents the common problem of near blindness of the conventional reference based method along edges of an image due to variations in the image.