Camera Test: PPF Needed For IDs, Text, Money

Avatar
Derek Ward
Published Feb 19, 2014 05:00 AM

Need to see the fine print of a dollar bill, euro, driver's license or text on a document?

We tested 3 different types of print.

Currency, both the US dollar and the Euro:

Identification card, a US state driver's license

And text, with point size ranging from 8 to 40, like so:

The tests were done with an overhead shot (like for cash counting):

And an angled shot (like at a bank teller):

This test answers the questions of:

  • What resolution do I need?
  • How wide of a FoV can I cover?
  • What impact do angles have on capturing this?
  • How many PPF do I ned for text, money and IDs?
  • Users seeking best detail of papers, currency, IDs, or other objects are best served by angling the camera, instead of looking straight down from above. Where this is not possible (due to obstructions or existing cable pathways, etc.) lights should be as diffuse as possible to reduce reflectance issues. Finally, WDR cameras may reduce some light handling issues, mitigating overexposure problems.

    The chart below provides our PPF recommendations for capturing currency, IDs, and other objects in this test:

    Notable Findings

    Here are our key findings from this test:

    • Placing cameras directly above the target (such as cash registers) may result in more light reflecting directly into the camera, overexposing the image.
    • Using an angled view of ~45° reduces reflectance issues while still providing similar details in most cases. Only the driver's license in our test was more difficult to read when the cameras were angled, due to its construction. 
    • Credit card numbers are especially difficult to discern at any resolution and FOV, since most modern credit cards use solid colors for both numbers and background to protect card holder information. Even at 1000+ PPF, numbers are only partially legible.
    • Lux levels on target for objects lying on a table or counter are much higher than expected compared to subjects elsewhere in the room, since more overhead light is reflected onto these objects. For example, subjects in our test environment are typically illuminated to ~160 lux on target, while objects facing up read 700+.

    Overhead Image Comparisons

    The image comparisons below complie overhead shots with the cameras mounted directly above the target, varying the horizontal field of view at 3', 5', and 9'. We take a look at varying font sizes, a Pennsylvania drivers license, €5 note, and $20 USD dollar bill.

    This image shows an overview of the scene at these FOVs:

    Varying Font Sizes

    We begin with a 3' wide field of view. In this case, 720p was only able to reliably discern 14 pt. font and above, while 12 pt. was legible in our 1080p camera. 5 and 10 MP cameras were able to discern 10 and 9 pt. fonts, respectively.

    Widening the FOV to 5', the 720p camera is only able to make out our largest font, 40 pt, and 1080p is barely able to discern 20 pt. At this FOV and light level, light handling begins to become an issue for the 5 and 10 MP cameras, with text smaller than 20 pt. impossible to read due to overexposure of the white page.

    Finally, at a 9' HFOV, both 720p and 1080p are able to discern only 40 pt. font, while the 5 and 10 MP cameras are so washed out no text is recognizable.

    Drivers License

    At a 3' HFOV, 720p and 1080p resolutions are unable to make out any of the text details of the driver's license, and lack detail in the subject's photo. Both the 5 and 10 megapixel cameras are able to read address, license number, and some personal details (greyed out below), and provide better detail of the subject photo.

    Moving to 5' FOV, none of the cameras are able to read text on the license, with overexposure becoming an issue due to the high reflectence of the plastic license.

    Finally, at 9', details are entirely lacking.

    Euro

    Looking at a €5 note, we can distinguish the denomination at all resolutions, but serial number (seen top center) is not possible to read.

     

    At 5' FOV, the €5 denomination is still easy to make out on all cameras.

    Finally, widening to 9', denomination is still visible, though with difficulty in the 720p camera. The 5 and 10MP cameras become overexposed here, but €5 is still visible.

    $20 Dollar Bill

    Looking a US $20 bill, all cameras are able to discern bill value and serial number at a 3' HFOV.

    Moving to 5' HFOV, the 720p and 1080p cameras no longer can make out the serial number due to lowered resolution. The 5 and 10MP cameras provide enough PPF for legibility, but exposure problems start to arise.

    Finally, at a 9' HFOV, the serial number is obscured in all cameras. The 10MP camera is so overexposed at this point that even the $20 mark is difficult to read.

    45° Viewing Angle Image Comparisons

    We moved our cameras back from the table and angled them at about °45 to see what effects this had on text legibility and exposure issues seen in our overhead shots. Again, we take a look at varying font sizes, a Pennsylvania driver's license, €5 note, and $20 USD dollar bill, seen in this FOV:

    Font Sizes

    At 3', details are essentially the same as the overhead shot, with 720p and 1080p discerning 12 and 14 pt. font, respectively. The 5MP camera is able to read down to 9 pt., and 10MP down to 10 pt. 

    Widening the FOV to 5', only 40 pt. font is legible at 720p, with 20 pt. visible in the 1080p camera. Overexposure in the 10MP image is modestly reduced here, with 14 pt. font marginally visible (previously 20 pt. was the smallest visible).

    Finally, at a 9' HFOV, overexposure is once again a problem for the 5 and 10 MP cameras, with fonts smaller than 40 pt. illegible.

    Drivers License

    The driver's license becomes more difficult to read when the caemras are angled. At a 3' HFOV, even 5 and 10 MP have difficulty discerning the largest letters on this ID.

    Widening the FOV to 5', details are near completely lost.

    Finally, widening the FOV to 9', no details are discernible, with the state barely legible in even the 10MP camera.

    Euro

    At a 3' HFOV, the €5 note is still easily visible in all cameras.

    Exposure issues are moderately reduced on the Euro, seen in this comparison. All cameras are still able to make out the denomination without issue.

    Finally, where the Euro was previously so overexposed the 10 MP camera had difficulty discerning the €5 mark, angling the cameras reduces this, making details of the note fully visible.

    $20 Dollar Bill

    Using a 3' HFOV, bill denomination and serial number is visible on all cameras with no issues.

    At a 5' wide HFOV, the serial number is partially visible in only the 10 MP camera, with others unable to make it out. Bill value is clearly visible in all cameras.

    Finally, at 9' wide, serial number is impossible to discern. However, exposure issues are reduced due to camera angle, making the bill denomination visible in all cameras.

Comments (13)
JC
Jason Clement
Feb 19, 2014

Fantastic test, this great information!

CC
Chris C
Feb 20, 2014

Thanks for this test guys. This is very helpful!

JH
John Honovich
Feb 20, 2014
IPVM

Jason, Chris, thanks for the feedback!

Everyone else, especially those who complain about IPVM being controversial, this has 80% less reads than our 'controversial' / business trends pieces. If you want more hard core tests, read them and show that you appreciate them.

JH
Jim Hall
Feb 20, 2014

I read'em all no matter how juicy or how dry...

Reason is: I ain't missin' no questions on the next pop quiz like I did on the last one...

Which imma guessin is any day now...

HL
Horace Lasell
Feb 21, 2014

Sorry, I had edited this before I saw your reply. I had asked if my prior comment was available for display. There was a posting error that presented an incomplete report, which I had commented on. Beyond completeness issues, I had basically expressed appreciation of the effort and asked about the possibility of OCRing the text images.

After I loaded the 3ft_Font_Sizes.jpg text image into Acrobat to OCR, I read...

"Acrobat could not perform recognition (OCR) on this page because:

The resolution of this page is below the minimum 144 dpi.

OK"

I was surprised that the image carried sufficient metadata for Acrobat to have any sense of its true dimensions, and would not even attempt to OCR it.

For example, when I tried to OCR images of ships with 12' high letters across the stern, Acrobat at least attempted to OCR the images (without success).

JH
John Honovich
Feb 21, 2014
IPVM

It was deleted because your comment was on a non published, non complete report.

RT
Ryan Talbot
Feb 21, 2014

Thank you. Definitely a well rounded series of tests.

Clearly the 1080p is the overall winner here when it comes to VFM. The 10mp is not 5 times better than the 1080p or is the 5MP 2.5 times better than the 1080p.

What is interesting is the color reproduction between the two HD and the MP images.

Could you please confirm if the screen shots were all the same compression format i.e. MJPEG or H.264 and if it was done through the cameras native web interface or through a VMS.

JH
John Honovich
Feb 21, 2014
IPVM

Ryan, we always test with H.264, unless otherwise noticed, given that is the overwhelming codec of choice in the industry. I'll let Ethan answer the other questions.

RT
Ryan Talbot
Feb 21, 2014
Ethan, Could you please advise the method of capturing each image. Was it through a specific VMS or through the cameras native web browser. Were they the same 4 cameras for all of the tests. What was the image quality settings and H.264 profile used.
ER
Eldon Rollins
Feb 21, 2014

I appreciate this report as much as any of the others IPVM provides; without comparison tests like these, we'd be performing these tests on our Customers....not the best scenario.

Thanks for the information.

DB
Doug Belt
Feb 25, 2014
IPVMU Certified

I thought this was a good test showing how the different "documents" were influenced by the overhead light at the two different angles utilized. Good food for thought if one is faced with trying to get a "best" of both worlds for overhead/casino type images of a variety of "documents" which may come across a counter or desk.

DS
Don Soper
Aug 16, 2017

Two Questions Please, John:

If I missed it, I'm sorry, but do you have a recommendation on the range of PPF for reading license plates?

Do you know if the PPF numbers would vary with H.265?  I saw some chatter at IPVM on this topic last year, but it is more of a valid question now than it was in 2016.

Thank you!

JH
John Honovich
Aug 16, 2017
IPVM

Don,

Good questions.

License plate PPF requirements are similar to human face ones - 50ppf+ in general. You can see that / experiment with that in our Calculator example.

The big issues with license plate are (1) dealing with lighting (headlights + nighttime, etc.) and (2) dealing with speed (i.e., how fast is the car moving). This is why LPC generally requires integrated IR + fast shutter speeds, something we review in this license plate capture test. Regardless of your PPF, if lightning issues and speed are not properly handled, the license plate won't be properly captured.

Do you know if the PPF numbers would vary with H.265?

Nothing fundamentally makes H.265 better or worse than H.264 for capturing license plates. How much one compresses the image does, irregardless of the codec (e.g., see our Video Quality / Compression Tutorial).

(1)