On Tuesdays the USPTO issues new patents. Today's Spotlight Patents concern digital fingerprinting and digital watermarking. Assigned to Microsoft, the first patent addresses techniques for forensic for fingerprint detection in multimedia. Assigned to Digimarc, the second patent addresses techniques for reversable watermarking.
7,562,228, "Forensic for fingerprint detection in multimedia," assigned to Microsoft.
The subject matter includes systems, engines, methods, and schemata for embedding spread-spectrum fingerprints in multimedia content: a first fingerprint to identify a recipient of the multimedia content and a second fingerprint at a location in the first fingerprint that represents a subset of a much larger group of recipients. By locating the second fingerprint, the search for a media pirate is immediately narrowed down to the subset, thus immensely speeding up pirate identification. In one implementation, chips of the second fingerprint can be made small and embedded sparsely, making alteration difficult. Systems, engines, methods, and schemata for synchronizing a pirated copy with original multimedia content to facilitate fingerprint recovery are also described.
7,561,714, "Reversible watermarking," assigned to Digimarc.
A reversible watermarking method embeds auxiliary data into a data set, such as an image, audio, video or other data, in a manner that enables full recovery of the original, un-modified data set. This method may be used to determine whether the data set has been tampered. To improve embedding capacity without the need for compression of the auxiliary data, the method uses an expansion technique. One particular approach exploits the correlation or redundancy within the data set to convert the data to a set of small, expandable values, such as difference values. These small values are then expanded by inserting auxiliary data as one or more additional bits, increasing the number of bits without causing an underflow or overflow. This approach also uses a property of the data set that is invariant to the embedding operation to identify embedding locations, obviating the need for separate data to identify where data is embedded in a data set.