Stamp Image Recognition

Hi all,

I am new to this forum and a relative newcomer to the hobby. My occupation is programming and since I was not able to find an app for quickly identifying stamps from a picture in order to add them to a digital collection I have created one on my own.
Is there an interest in such an app? *** Edited by Staff - Please Review the rules that you agreed to when you registered. *** I know such posts are frowned upon in some communities and I do not have the necessary post count yet.

Welcome,
Since we do not know you and installing apps from unknown sources can be problematic, we err on the side caution. Please contribute to the community for 50 posts before soliciting.
Don

I understand. I would still like to gauge the community's interest in such an app and features the members would like it to have.

Understood, getting this kind of feedback is fine.
Don

I am interested in learning from you what you "app" exactly does to contribute to our hobby as stamp collectors. Just a description I am sure would be accepted.....

After taking a picture of a stamp, in a few seconds, the user is taken to its corresponding page on a popular collectors website. Info about the stamp is listed on the page, it can be marked as in one's collection or as up for trade. Also, it can be checked who wants the stamp or can trade it, among other functionality.

It can help with stamps one is having trouble to identify. I personally use it to greatly speed up the process of adding my stamps to my "digital collection" on the website. Perhaps it is geared towards beginner collectors more, I would like your feedback to determine this.

Isn't it expensive for you to get the image sources (rights to compare to them) of all those stamps?

I think (17-06-18) Not Allowed - Auto-Removed (function called "SRS") does this recognition also, but I don't use this software, so didn't test it.

---
(still don't know why it is not allowed to name stamp software titles here?)

As a fellow geek I'm more than interested on the inner works. What kind of approach does it use for image recognition, and what kind of limitations there are to it. If I take a snapshot with 16 cameraphone, can the 'app' process it or will it choke (does it process them locally or on cloud/server). What if the stamp is 'warped' in image, or rotated/upside down? OOr image has a poor focus? Or how does it fair with stamps which have got obtrusive postmark or poor contrast? Or... So a proper technical description on what it does/how it works (beyond take an image + put it into app) would be a fairly good start. What kind of factors does it look from image, what kind of image is the app looking for, how accurate is the comparison etc etc.

The idea in itself isn't new, there are number of commercial products (and couple of university student projects which are basically free/opensource)... So far most of the options I've seen have been more or less handicapped (partially because the comparison images are of low/varying quality, partially for number of other reasons).

-k-

I have always been interested in such technology specific to plating the 1 cent Franklins. Now THAT would be nice!

I don't think you actually realize how many stamps and colors and perforations exist. Though this would be a great idea, it's virtually impossible for an image recognition software for this subject.

Issues with stamp recognition have been with us for about 10-15 years and they have plagued all the previous attempts.

The first challenge are black and grey scale stamps; how does the software differentiate a blank cancel from the actual stamp design?

But the biggest challenge is the ability to recognize the proper stamp when perforations and watermarks make or break the identification. Can software make basic recognition based upon a stamp design? Certainly. Can it determine watermarks, paper types, tagging, or printing methods? No. And unfortunately these things make a big difference in IDing many stamps.
Don

@stamperix
The website has granted me rights to use the images for the purposes of this app. I suppose you are referring to a desktop app, I know of a few that do image recognition, but none on mobile. (I could be wrong though)

@scb
It is based on ORB descriptors, which deals reasonably with rotation, contrast, shadows, etc. It does work with light cancelations, but heavy ones are a problem (or when the indexed image has a cancel). The image is processed on a server, as there is a high memory requirement to keep the index of all image hashes in RAM for the processing to be fast.
I do not have proper statistics on how often it is successful as I have not implemented any tracking nor do I yet have the users to gather any meaningful data. I can share that in my own use it does manage to identify the stamp 80% of the time. I am probably biased though :) Low-quality source images have been ignored in the indexing process, the index currently holds ~510 000 stamps from all countries from 1957-now because of server resources limitations.

@51studebaker
You are correct, it is based purely on stamp design, maybe it is better geared towards beginners like me? It does have 510,000 images indexes as I said in the previous post, so I think that is a fair amount.

I am glad that it does generate some interest, I still wish I was allowed to share the app itself, so you could give it a go, not take my word for it.

Technically speaking it would very much possible, but it doesn't happen without certain (technnical and non-technical) requirements. A good example of such large scale solutions are facial recognition systems used by Facebook and alike.

-k-

I've come across with it when I've played with Opencv and other computer vision libraries ;)




Makes perfect sense, but what worries me most here is the networking side of things. Even if you could serve fairly many simultaneous users on the server, there is the question on how end users perceive the general network latency on uploads (which is basically something you don't control and where even CDN doesn't help).

And of course your bandwidth consumption could/would be pretty astronomical - which takes me to question how would you fund/monetize it.

-k-

Of course, delay while uploading is inevitable. As for bandwidth and simultaneous connections, I am prepared to handle the number of users I expect to get - this is more of an niche app rather than something that appeals to everyone.