The first thing we do

A pattern of obfuscation may be just what the lawyer ordered when it comes to patents and machine vision.

In 2004, after years of litigation, Cognex (Natick, MA, USA; www.cognex.com) announced that Chief Judge Philip Pro of the US District Court in Las Vegas, NV, ruled in favor of the company in its lawsuit against the Lemelson Medical, Education & Research Foundation, Limited Partnership. As you may recall, the Lemelson Foundation (Portland, OR, USA; www.lemelson.org) claimed that it held multiple patents related to machine vision without actually building a single device or system. Luckily, Judge Pro held that the claims of the 14 patents were invalid, unenforceable, and not infringed by Cognex.

In the four years that have passed, however, it seems that little has been learned from this landmark ruling. Reviewing many of the patents currently listed at the United States Patent and Trademark Office (USPTO; www.uspto.gov) reveals the numerous flaws that have yet to be addressed in the patent applications that describe both the development of technology and the deployment of systems that use image-processing or machine-vision components.

On the technology front, multiple patents are listed that describe image-processing algorithms. One patent, entitled “Adaptive edge detection and enhancement for image processing,” describes just that. Upon careful examination of this patent, however, one has to question what is really original about the claims, since the patent is surrounded in legalese. Interestingly, too, the patent references numerous classic works such as Azriel Rosenfeld’s book Digital Picture Processing, written nearly 20 years earlier, as well as conference proceedings and IEEE papers.

While legal mumbo-jumbo may be standard procedure for the patent office, it certainly opens itself to abuse, especially in any technical field. Edge-detection algorithms, for example, are one of the most commonly used algorithms in machine vision. But does anyone claim a patent on, for example, subpixel interpolation using cubic splines? I could not find any such reference to this at the USPTO.

If I were an unscrupulous person, I would submit such a patent immediately. And the reason would be, of course, to make lots of money. If I held such a patent and knew that certain commercial software used this subpixel interpolation method for edge detection, would I then have the right to claim patent infringement?

Unfortunately, the situation does not just end with those obtaining patents for algorithms. On the systems front, as you will discover, there are some companies that have patents that describe complete systems that use the technology. Upon examination of these “original systems,” it is also often difficult to understand what is original and why the patent was issued in the first place.

Originally, the role of the patent office was to protect the intellectual property of the developer of a unique product. Now, however, the rapid dissemination of information that is possible using the Internet has blurred the distinction between original thought and derivative claims.

Patents needed?

All of this leaves the software and system developer with little choice. If an original algorithm or system has been invented, then a patent should be obtained. Unfortunately, one has to wonder whether any patent attorney has ever read the IEEE Transactions on Pattern Matching and Machine Intelligence (www.computer.org/tpami). And, since the answer to this question is rather obvious, developers of original products tend to avoid providing any type of in-depth technical descriptions to the patent office (or to the press), favoring instead to promote their advantages and benefits.

In Shakespeare’s King Henry the Sixth, Part II, Dick, the Butcher, remarks to Jack Cade, the Rebel, “The first thing we do, let’s kill all the lawyers.” If he were alive today, Shakespeare may have written about technology, and I think he would have chosen the word “avoid” rather than “kill.”