How necessary are 1000 PPI Images?

[Mike French]

Recently a smaller LiveScan vendor explained the costs involved developing 1000 PPI LiveScan vs 500 PPI. I learned that the cost becomes significant especially to a small vendor competing with industry giants on contracts with 1000 PPI requirements.

The question from the vendor was whether 1000 PPI really gave that much more information to the examiner, and I went through my usual explanation that it has everything to do with examiners and nothing to do with AFIS.

Today I listened in on NIST’s 60th anniversary in biometrics and the same conversation came up and the panel really didn’t have more of an answer than I just gave.

I remember when I first became an LPE and @orrb did an informal survey in the office, and generally found that 600 PPI satisfied our latent comparison expectations. I’m wondering if anyone has seen studies specifically addressing the 500 vs 1000 PPI difference.

Anyone have strong feelings about this?

[Bill Schade]

These debates reminds me of the benchmark studies on AFIS accuracy when the differences between systems came down to micro percentage points. Other than providing bragging rights to vendors, did they really help customers make good decisions?

I think a clear 500 ppi image is alot more valuable than a smudged 1000 ppi image.

Clarity made the job easier (and improved accuracy) rather than resolution.

But then again, I've always been a practial person

Bill Schade

[ER]

By math, a 1000ppi image contains 4x the pixels of a 500ppi image.

Practically, 1000ppi images result in higher accuracy searches. Modern AFIS use the actual images in addition to the minutiae data and they additional data increases accuracy.

So the real question remains... Do you want to get a hit? Yes? Then get a 1000ppi database of tenprints and latent prints of resolution as high as possible.

[Steve Everist]

By math, a 1000ppi image contains 4x the pixels of a 500ppi image.

Practically, 1000ppi images result in higher accuracy searches. Modern AFIS use the actual images in addition to the minutiae data and they additional data increases accuracy.

So the real question remains... Do you want to get a hit? Yes? Then get a 1000ppi database of tenprints and latent prints of resolution as high as possible.

Since you work for Idemia, has there been any testing to verify that the higher the resolution, the more likely to get a hit (all else the same)? Personally, as someone who started by printing with ink in a jail and comparing inked prints as an LPE, all else the same I'd rather have a known impression taken from direct transfer using ink. But that's not the reality we live in anymore. That being said, is there a point at which additional resolution is no longer a benefit and just a waste of drive space?

Also, I've always been fascinated with how such round numbers of 500ppi and 1000ppi have come to be. Maybe George Reis can stop by and clarify this, but from a flatbed scanner perspective, they're natively 600ppi and 1200ppi, so the scans are doing the same pass but also throwing out resolution. It seems like the standard-setters were just going for round numbers. Maybe this is different in livescan machines - I don't know.

[Mike French]

A little more background, the vendor that brought me the problem described 1000 ppi LiveScans 10x more expensive to develop than 500 ppi. And wondered the cost benefit...

I've seen at least two smaller studies where downsampled images performed better than their higher resolution counterparts, on some AFIS tests. Presumably because the downsampling reduced noise which interfered with the feature extractor. Which got me thinking about this topic.

However, its entirely possible that Deep Learning models in the near future will make better use of the increased number of pixels. As a good friend just pointed out to me in an email. I can't argue with that logic either. So I can admit I have no strong feelings on the issue

[Steve Everist]

Another thing not being discussed is the affect resolution (or lack thereof) when using 500ppi vs 1000ppi images for the comparison process. And, has livescan and the digitization of known cards rendered the reliability of doing comparisons to pores, ridge shapes, edges, etc... unreliable? And, if not, at what resolution does it become so?

[antonroland]

I have a question.

Where did the 500ppi thing come from from when the whole camera and scanner industry built itself around multiples of 300 dpi?

[Steve Everist]

I have a question.

Where did the 500ppi thing come from from when the whole camera and scanner industry built itself around multiples of 300 dpi?

Without knowing the actual answer, I'd guess by a group of people who create policy without knowing process.

[antonroland]

I have a question.

Where did the 500ppi thing come from from when the whole camera and scanner industry built itself around multiples of 300 dpi?

Without knowing the actual answer, I'd guess by a group of people who create policy without knowing process.

Hi Steve

Yep, THAT sounds very familiar...