Rare Coin Grading System

The goal of the project is to develop an automated system that will be used to identify and grade rare coins, providing consistent and repeatable results. Rare coins are presently graded by human hand and eye inspection that often produces varied, inconsistent, and sometimes dubious results. A difference of a single grade can often mean thousands of dollars in the value of the asset. Judgment is suspect with subjectivity and great financial incentives entrenched in the process.

The team will develop a system that is capable of extracting features and conditional grade information from scanned (front and reverse) images of United States Business strike coins. Ideally, the developed system should be able to properly identify the following from a scanned image:

  1. Series (example: Lincoln Cent, Indian Cent, Jefferson Nickel, Roosevelt Dime, etc.)
  2. Denomination (cent, 5 cent, 10 cent, 25 cent, 50 cent, $1)
  3. Year (for Lincoln Cents: 1909-current year)
  4. Mintmark (for Lincoln Cents: D=Denver, S=San Francisco, none=Philadelphia)
  5. Grade (fair, about good, good, very good, fine, very fine, extremely fine, almost uncirculated, and uncirculated)
The system should to be trained with several hundred sample coins of various denominations, series, grades and mintmarks. The client of this project will provide coins required for training and testing the system.

To simplify the problem we will limit the scope of this project to Lincoln Cents to fix the series and denomination of the coins and to limit the pattern classification task to the year, mintmark, and grade.

Project tasks and timeline milestones:

  1. Gain Domain Knowledge all team members should have a full understanding of the terminology, tasks required, and technology required.
  2. Establish Image Standards determine the scanner (or camera), the software, and the resolution to be used for capturing and saving images.
  3. Build Expert Visual Database build a database of images of 300400 pre-graded coins for subsequent training and testing.
  4. Develop the pattern classification software to recognize the date and mintmark of a coin.
  5. Develop appropriate features and pattern matching software for grading the coin this is the heart of the system as it involves the software to compare the features of the coin to be graded to those of pre-graded coins in the database. The system should now return the year, mintmark, and grade of the coin.
  6. Bootstrapping each time a user-supplied coin is successfully identified and graded it should be added to the database as accumulated knowledge.
  7. Develop an intuitive web-based GUI that allows users to submit a scanned image to the system. This GUI should return the year, mintmark, and grade to the user.
  8. Tie in obtained results to a values database set up a values database (weekly values are available on the Internet) and obtain the value of a scanned coin from the series, denomination, year, mintmark, and grade.
A prototype system should be completed by the end of the first semester (tasks 1-5) and a robust and complete (remaining tasks) one by the end of the second.

A team last spring from CS631Q Pervasive Computing Systems developed a preliminary system that took a scanned image of an input coin and obtained Hue (H), Saturation (S), and Brightness (B) histogram vectors for comparison the HSB vectors of the pre-graded coins to arrive at a grade for the input coin. The preliminary results under certain conditions were quite good, although some scanned images resulted in serious grading errors. Complete documentation of this work will be available to the CS615 team that undertakes this new project.

Richard Bassett is the client for this project and he expects to work closely with the team that elects this project. He is a Professor at Western CT State University and a DPS student at Pace. Components of this project will be used in his DPS dissertation.