Computer-Assisted Coin Grading

Well, they are directly infringing on a patent which PCGS has held for almost twenty years. I don't see it lasting very long.

SD, I'm not an expert in patent law nor do I know the specifics of PCGS's patent. But as an over-educated layperson and technologist, I would be shocked, SHOCKED, if PCGS had patented the entire concept of computerized coin grading rather than just the particulars of their chosen process.

I know a little about image processing; computer processing capabilities and algorithm sophistication have made considerable advances in the past 20 years. If a company can't persuade the patent office that they are doing something that hadn't been considered 20 years ago, then the patent system is seriously flawed.

SD,

I spent some time with the patent you linked to. The following describes how I understand its contents:

quote:

---

A. Reference database generation

(1) for a given coin type, collect thousands of examples spanning all imaginable grades
(2) expertly/manually define a set of features (grade-determinative characteristics) of said type
(3) expertly/manually assign a grade to each coin in the database

B. Digital capture process

(4) digitize each member of the reference database [NB, this is the imaging step]
(5) for each feature from (2), develop a mapping from the image chip (4) to a discrete value or set of values (aka, feature vector) [NB, this is the image processing step]

C. Grading algorithm development

Now, divide the reference database into training and testing samples.

(6) using the training samples, employ a computer algorithm to determine the optimal weight for each of the feature variables from (5) in order to maximize recovery of the grades assigned in (3) [NB, this is the computer learning step]
(7) with the weights derived in (6), repeat the process on the testing samples, again with the goal of recovering the grades assigned in (3) [NB, this is the validation step]

---

If the results are satisfactory, great, you have your system. If not, go back to step (6) and rerun your algorithm using different initial conditions. Still no success--modify the computer algorithm [the patent referred to linear regression and a genetic algorithm, but there are other choices]. Still no success--go back to step (5), where there are many choices to be made. Still no success? perhaps steps (4) or (2) need to be re-addressed.

I went through this exercise to illustrate a couple of points:

• First of all, my description of the process was a lot more general than the description in the patent. I assume PCGS made their description as general as they could get away with. Consequently anywhere my process differed from their process, there is an opportunity for a competing party.


• Although PCGS has/had an advantage in part A, I wonder how many heavy-hitters they hired to develop parts B and C.


• How do you perform the digital capture? Hi-res scanning, photography, or 3-D laser morphology? Color or monochromatic? Does the PCGS pattent specify their method or preclude others?


• There are so many choices to make in steps (5) & (6), it boggles the mind. There are academic fields devoted to them, in fact. If you assume that the relationship between the characteristic features and the final grade is too complex to readily codify, it is natural to rely on a genetic algorithm or neural net to do the heavy lifting for you (i.e., determine the variable weights). But that need not be the case. You have the experts, go full-out creating an expert system (i.e., have the graders explain exactly how they interpret each of the characteristic features and how they put it all together to form a grade--and codify that).


• There is still another, simpler approach. After you complete step (4), create an electronic photgrade system: For each coin to be graded, use image matching to electronically score its agreement with each of the exemplars from the database. Let each "vote" and you find out which grade the sample best matches. This is essentially what you do when you use photograde to grade your coins. Only the electronic system would draw on thousands of reference images where you have only a handful.

In summary, if I were consulted (as a technologist, not as a legal expert), it would be easy for me to testify that there is immense room for other innovators in the field of computerized grading.