| US 7,574,050 B2 | ||
| Image page search for arbitrary textual information | ||
| Michael Ladwig, Reston, Va. (US); Bruce W. Stalcup, Springfield, Va. (US); and James Brower, Ashburn, Va. (US) | ||
| Assigned to Northrop Grumman Corporation, Los Angeles, Calif. (US) | ||
| Filed on Mar. 10, 2004, as Appl. No. 10/795,976. | ||
| Application 10/795976 is a continuation in part of application No. 09/126818, filed on Jul. 31, 1998, granted, now 6,741,743. | ||
| Prior Publication US 2004/0170328 A1, Sep. 02, 2004 | ||
| Int. Cl. G06K 9/62 (2006.01) | ||
| U.S. Cl. 382—209 [382/280] | 16 Claims |
| 1. A method of automatically identifying a pattern on a page, comprising:
a computer performing the following:
synthetically generating textual patterns as signal templates;
generating a compensated image, compensating, if necessary, for visual differences between the synthetically generated textual
patterns and images being compared against the synthetically generated patterns; and comparing compensated images against
images in a database; and
identifying a pattern on a page based upon the comparison;
wherein compensating an image includes reducing resolution, inverting and mirroring a page image in the database; and
wherein comparing compensated images against images in a database includes:
performing a Fast Fourier Transform (FFT) on the compensated image, moving the compensated image from the spatial to frequency
domain;
performing an FFT on a search target image;
multiplying the FFT on the image to be searched by the FFT of the search target image to produce a correlation plane;
performing an inverse FFT of the correlation plane, converting the correlation plane from the frequency domain to a spatial
specification to produce a similarity matrix for search pattern locations within the target image; and
applying a threshold to the similarity matrix to extract location of matches above the threshold.
|