Abstract: A template image of a form document is processed to identify alignment regions used to match a query image to the template image and processing regions to identify areas from which information is to be extracted. Processing of the template image includes the identification of a set of meaningful template vectors. A query image is processed to determine meaningful query vectors. The meaningful template vectors are compared with the meaningful query vectors to determine whether the format of the template and query images match. Upon achievement of a match, a homography between the images is determined. In the event a homography threshold has been met, information is extracted from the query image.

Abstract: The present invention discloses a method for comparing a camera fingerprint to a query fingerprint. An estimate of a camera fingerprint is obtained from a set of one or more images, and the query fingerprint is obtained from one or more images. Using these values, normalized cross-correlations values are determined for each possible alignment of the two fingerprints. Prior to calculating the noise floor, a set including the highest normalized cross-correlation values is identified. A universal noise floor is then calculated excluding this set from the noise floor calculation. The universal noise floor is utilized in calculating a correlation energy for each possible shift. The correlation energy values are then examined to determine whether the camera fingerprint and the query fingerprint match. A visualization tool may also be used to compare the camera fingerprint and the query fingerprint and determine whether the camera fingerprint and the query fingerprint match.