Counterfeit detection

[realgoods]

I’m assembling a proof of concept anti-counterfeiting system involving microscopy. Imagine there exists a system which compares digital images of objects (currency, jewelry, etc) against known counterfeit and authentic examples and provides a probability of authenticity.

I have very little experience with microscopy (background is in software) but my hunch is there are surface microscopic differences (due to differences in the manufacturing process) between authentic and counterfeit objects.

What qualities should I be looking for in a microscope setup to consistently image the surfaces of such objects? I know it must support USB-C so it can be connected to a device as a webcam (video). The system need not record the video in case that matters. Per Microbe Hunter’s insightful videos, I‘ve learned that magnification is mostly market speak and other criteria are what really matters.

Thanks in advance!

[Scarodactyl]

Most dslrs will connect via usb as well, and you will get more bang for your buck in many cases.
It would help to know how you want to illuminte the subject (thus what kind of lighting you need), and what field or fields of view you need.

An indistrial zoom lens like a navitar, optem, unitron, or monozoom 7 would probably be a good starting point though. They have a long working distance and a range of magnifications. They will directly accept a c mount microscope camera, or sometimes you can get a dslr adapter though they are kind of rare. And systems can often be had very affordably.

Knowing your approximate budget would help too.

[PeteM]

FWIW, forensic comparison microscopes often show up used. These are basically two microscope bodies side by side, with a bridging head and prism that allows two near-identical images to be superimposed. Then something like the markings on two bullets (or two typewritten characters) can be compared to see if they came out of the same barrel (or from the same typewriter). I'll sometimes drag one of these to one of our "Micronaut" programs. The thing to know here is that you'll need stages that can present the articles in the same orientation. You'll likely need something similar for digital image recognition.

Most manufacturing processes will leave tell-tale markings which should be consistent if the items came out of the same mold, were finished with the same grit grinding wheel, followed the same CNC tool path etc. And manufacturers will often provide some sort of code (molded, laser engraved etc.) to determine the origin of a part from what plant down to what mold etc. Some automakers are even painting on "microdots" with a VIN code for traceability. It gets harder when you try to generalize across many different production processes and object types and they have been subject to wear and tear.

Don't know your programming chops, but an AI application where you fed the imaging systems thousands of known good and hundreds of known fake objects might be the best bet at creating a generic visual authentication system?

There's also a long history of analog video and then digital inspection systems for quality control. Likely plenty of prior knowledge available there. And today's forensic comparison systems are digital as well, though far as I know it's usually a trained human doing the comparison.

[realgoods]

Most dslrs will connect via usb as well, and you will get more bang for your buck in many cases.
It would help to know how you want to illuminte the subject (thus what kind of lighting you need), and what field or fields of view you need.

An indistrial zoom lens like a navitar, optem, unitron, or monozoom 7 would probably be a good starting point though. They have a long working distance and a range of magnifications. They will directly accept a c mount microscope camera, or sometimes you can get a dslr adapter though they are kind of rare. And systems can often be had very affordably.

Knowing your approximate budget would help too.

Scarodactyl -- thanks for you detailed response. Advice outside this forum confirmed your point that I may not even need a microscope for this (perhaps a zoomed DSLR as you suggest). They mentioned the need for "high-intensity" light -- though not exactly sure what that means other than as bright a light as I can find? Budget for the proof of concept is <= ~1K or so. Although if it shows promise budget would definitely be expanded. Basically, I'd like to show that the idea works in principle.

[realgoods]

FWIW, forensic comparison microscopes often show up used. These are basically two microscope bodies side by side, with a bridging head and prism that allows two near-identical images to be superimposed. Then something like the markings on two bullets (or two typewritten characters) can be compared to see if they came out of the same barrel (or from the same typewriter). I'll sometimes drag one of these to one of our "Micronaut" programs. The thing to know here is that you'll need stages that can present the articles in the same orientation. You'll likely need something similar for digital image recognition.

Most manufacturing processes will leave tell-tale markings which should be consistent if the items came out of the same mold, were finished with the same grit grinding wheel, followed the same CNC tool path etc. And manufacturers will often provide some sort of code (molded, laser engraved etc.) to determine the origin of a part from what plant down to what mold etc. Some automakers are even painting on "microdots" with a VIN code for traceability. It gets harder when you try to generalize across many different production processes and object types and they have been subject to wear and tear.

Don't know your programming chops, but an AI application where you fed the imaging systems thousands of known good and hundreds of known fake objects might be the best bet at creating a generic visual authentication system?

There's also a long history of analog video and then digital inspection systems for quality control. Likely plenty of prior knowledge available there. And today's forensic comparison systems are digital as well, though far as I know it's usually a trained human doing the comparison.

PeteM, thanks, spot on --- the plan is to use an image classification model (AI / ML) to do the deciding. Human's will be in the loop but as a quality control layer.