(from Patent abstract)
The present invention uses invisible junctions which are a set of local features unique to every page of the electronic document to match the captured image to a part of an electronic document. The present invention includes: an image capture device, a feature extraction and recognition system and database. When an electronic document is printed, the feature extraction and recognition system captures an image of the document page. The features in the captured image are then extracted, indexed and stored in the database. Given a query image, usually a small patch of some document page captured by a low resolution image capture device, the features in the query image are extracted and compared against those stored in the database to identify the query image. The present invention also includes methods for feature extraction, feature indexing, feature retrieval and geometric estimation.