The technique uses two methods to trace a document: first, by analyzing a document to identify characteristics that are unique for each printer, and second by designing printers to purposely embed individualized characteristics in documents.
<?xml:namespace prefix = o ns = “urn:schemas-microsoft-com:office:office” />
The technique currently focuses on laser printers but eventually will be expanded to inkjet printers, said Edward J. Delp, a professor of electrical and computer engineering at Purdue.
Findings of the research, funded by the National Science Foundation, will be detailed in three papers to be presented on Nov. 5 during the International
Conference on Digital Printing Technologies in <?xml:namespace prefix = st1 ns = “urn:schemas-microsoft-com:office:smarttags” />Salt Lake City.
The papers were written by Delp; Jan Allebach, a professor of electrical and computer engineering; George Chiu, a professor of mechanical engineering; and engineering doctoral students Pei-Ju Chiang, Gazi N. Ali and Aravind K. Mikkilineni.
Counterfeiters often digitally scan currency and then use color laser and inkjet printers to produce bogus bills. Forgers use the same methods to make fake passports and other documents.
“Investigators want to be able to determine that a fake bill or document was created on a certain brand and model of printer,” Delp said.
So far, the researchers have been able to successfully identify which model of printers was used to create certain documents in 11 out of 12 models tested, according to data to be released during the conference.
“We also believe that we will be able to identify not only which model printer was used but specifically which printer was used,” Delp said. “That means we will be able to tell the difference between counterfeit bills created on specific printers even if they are the same model.”
Officials also would be able to use the method to determine the authenticity of documents, such as airline boarding passes and passports. Such information would enable homeland security investigators to determine from what country or regions of the world certain printed documents originated, which could help trace the location of potential terrorists and their collaborators.
The technique uses specialized software to detect slight variations, or “intrinsic signatures,” of printed characters, revealing subtle differences from one printer to another. Even printers that are the same model have slight flaws and variations in their mechanical systems. These variations result in subtly different characters.
“We have observed variability from printer to printer within a single model, ” Allebach said. “That’s because for a company to make printers all behave exactly the same way would require tightening the manufacturing tolerances to the point where each printer would be too expensive for consumers.
“This is a very competitive market right now. You can buy a color laser printer for less than $500. One of the ways in which manufacturers are able to make printers so affordable is by cutting corners. The gears are made of plastic, and they are not made extremely accurately. There also is variability from toner cartridge to toner cartridge.
“We are able to determine this intrinsic signature based on knowledge of the physical characteristics of the printer mechanisms.”
Allebach and Chiu have been working with printer companies for more than five years to reduce a phenomenon called “banding,” which are horizontal imperfections in the print quality of documents.
“Banding arises whenever you have a print mechanism that uses rotating components,” Chiu said. “What happens is the components don’t necessarily rotate at an exactly constant speed.”
Inside of a laser printer’s cartridge is a “photoconductor drum,” which rotates as a laser beam scans back and forth along the drum. The drum is coated with a charged material that releases its charge when exposed to light. The laser turns on and off rapidly, selectively removing the charge in certain areas. Toner is attracted to those areas that no longer have a charge, forming letters or features of an image, which are then transferred onto sheets of paper.
“This process is called development,” Allebach said. “Because of variability in printers, the drum does not rotate at a constant speed. If the drum slows down a little bit as it is rotating, you get excessive development, so the print will look a little dark. And where the drum speeds up, you get too little development and the print will look a little bit light.”
The resulting bands of light and dark cause imperfections in a text document or an image. Because every printer has its own unique pattern of banding, or intrinsic signature, the imperfections can be exploited to trace a document to the printer on which it was created, Chiu said.
“We extract mathematical features, or measurements, from printed letters, then we use image analysis and pattern-recognition techniques to identify the printer,” Delp said.
If, however, the printer cartridge is changed after a document is printed, the document no longer can be traced to that printer. The Purdue researchers are overcoming that problem with software that causes a printer to embed its own unique “extrinsic signature” in a printed document, regardless of which printer cartridge is in a machine.
“We will actually modify the way the printer puts marks on the paper,” Chiu said. “This method is very difficult to get around because information about the internal workings of specific printers is not commonly available, even on the Internet.”
Banding can be altered from one printer to another by adjusting the laser intensity, how long each laser pulse lasts and the precise positioning of a small motor that steers the laser beam inside the printer.
Chiu and Allebach have pioneered methods to reduce banding. The same methods they have developed to reduce banding also can be used to artificially embed bands that are too fine to be seen by the unaided eye but can be detected with image-analysis techniques.
“We need to understand the human visual threshold for the signatures we put in so that the features are strong enough to be detected with image analysis methods but not by the human eye,” Chiu said.
