Case Study

How Stamp Recognition can save costly, unnecessary hours and help recover considerable losses in unpaid revenue

loss of revenue through a combination of insufficient postage and counterfeit stamps is a common irritation in our industry. suffering on-going frustrations with this problem, our client, an international postal and logistics organisation approached us to see whether prime vision could help alleviate this.

Letters and parcels franked with stamps increasingly fail to achieve correct values compared to their weight. Items are often franked with counterfeit stamps or sent with already cancelled stamps. Post is sent, particularly internationally, where the paid value of the stamp did not match the cost of the required destination. Systems are in place to attempt to combat these issues, but still, approximately 1% of sent items franked with stamps have insufficient or even worse, no value. These seemingly tiny amounts add up and in fact, cost significant international postal and logistics companies tens, sometimes hundreds of million per year; In an industry, with tight margins, these percentages could be the difference between profitability and loss.

WHAT CHALLENGES DID WE FACE?

In simple terms, the task may sound relatively easy. Locate the stamp, identify the stamp, assess its properties and compare these to the sent item. However, in reality, this presents a series of unique and complex problems that each require a blend of new techniques and new technologies to reach the solution.

First, the location of the stamp needs to be detected. However, how do you know what is a stamp from part of an address label or the packaging? Also, while we can presume the stamp may be top-right: How do we know which way is up? The hugely differing traits of sender’s addressing habits, together with the random, topsy-turvy journey through a mail-sortation system, necessitates a solution able to adapt to the individual circumstances of each item sent.

Once located, the stamp needs to be positively identified. Our customer has a growing database of almost 1000 stamps currently in circulation. Any one of them could be a match. While the design and values can vary considerably, some designs are very similar, with maybe only a slight colour change or a gradually updated silhouette of an ageing monarch. A well-loved, long-standing design could remain in place but the printed value increased over the years through inflation.

Lastly, the status of the stamp needs to be determined. How does the value and status of the stamp correlate to the sent items characteristics? Was our customer getting paid the correct amount, if any, for their services?

All this, and in a fraction of a second.

SO WHAT ARE THE BUSINESS BENEFITS?

The R&D team at Prime Vision began by breaking down the solution into four separate but integrated processes. Location. Recognition. Interrogation. Relevance.

Location is achieved through the proven technology of phosphorus detection. The solution can quickly locate a pre-impregnated stamp. The previously troublesome problem of stamp orientation was solved by the development team creating a new system that works regardless of the presented position.

Commencing the Recognition process, the system instantly makes an image comparison to either immediately identify the stamp; or set aside for further scrutiny should any anomalies be flagged. Should further investigation be needed, a more in-depth object recognition process is undertaken to differentiate between similar stamps, identify counterfeit or flag any other discrepancies.

Once the stamp is positively recognised and identified as valid, the face value is then compared to the weight of the item (supplied by the host sorting system) to ascertain correct postage paid.

Lastly, to ensure the solution is always relevant, Prime Vision has introduced a client configuration interface, allowing the stamp image database be quickly updated, effectively re-training the system as new stamps are released.

Leading the development was Senior Researcher Sjaak Koomen, “We are really excited by the overall solution, and particularly proud of our leaps in innovation in the object recognition process. Initial development led to an off-set between speed and capability. To achieve acceptable processing time, we had to sacrifice performance. Although the result was a great improvement on the previously available solutions and the client would have been delighted with these results, it caused some personal pain for
myself and members of the R&D team that it could be so-much-better. In the end, we decided it wasn’t acceptable and set about re-thinking the process”.

After looking at the various possibilities, the R&D took their inspiration from well know children game, “guess who?” The previous system already utilised a technique to break up each stamp into a series of unique minute patches of 40 pixels and then set about looking for a direct image match. The new method took advantage of these individual images to very quickly narrow down the possibilities.

Sjaak explains. “Rather than comparing each stamp against the complete database looking for a direct match, we realised we could very quickly dismiss which stamp it wasn’t. In very basic terms, instead of the program asking, “is this your stamp? No. Is this your stamp? No. OK. Is this your stamp? No.”. The new technique said, “is it red? Yes. Ok. Then it’s one of these three”.”

“By effectively creating a unique fingerprint for each stamp possibility only a tiny portion of the image would need to be processed. The small images patches could be easily identified and in a far more efficient way. Less processing – more accuracy! This innovation increased the speed by up to 20% while allowing us to perform deep searches still. Positive results leapt from 95% to 99%”.

Sjaak explains. “Rather than comparing each stamp against the complete database looking for a direct match, we realised we could very quickly dismiss which stamp it wasn’t. In very basic terms, instead of the program asking, “is this your stamp? No. Is this your stamp? No. OK. Is this your stamp? No.”. The new technique said, “is it red? Yes. Ok. Then it’s one of these three”.”

“By effectively creating a unique fingerprint for each stamp possibility only a tiny portion of the image would need to be processed. The small images patches could be easily identified and in a far more efficient way. Less processing – more accuracy! This innovation increased the speed by up to 20% while allowing us to perform deep searches still. Positive results leapt from 95% to 99%”.

WHAT ARE THE BUSINESS BENEFITS?

In an industry with narrow margins, it’s these small percentages that make the significant differences to profitability. The business case for the Prime Vision Stamp recognition system is simple. The solution can save costly, unnecessary person-hours and help recover considerable losses in unpaid revenue.

Even a smaller national Post or International Logistics company would nationally process more than 10 million letter and parcels per day. Based on this, just the step-change in performance from 95 to 99% between the Prime Vision solution and previously available solutions identifies an additional €2.6m per year in lost revenue; plus, a further saving of €350,000 per year in wasted working hours*. Impressive results based on 10m items a day. A large national Post would process over 50m.

Get a free consult

    Privacy Preferences
    When you visit our website, it may store information through your browser from specific services, usually in form of cookies. Here you can change your privacy preferences. Please note that blocking some types of cookies may impact your experience on our website and the services we offer.