Maker Pro
Maker Pro

Experiment with finger vein scanner

I wants to do some DIY to make finger vein scanner with serial CCD camera and some infrared light. I found this cool stuff, shows how its work. I want to implement this like fingerprint reader. How we can make it possible ?
how_fv_works.jpg

Image courtesy: Hitachi
 
Better what ?

scanning veins ? ( uses transmissive IR )
or finger scanner ? (uses refective imaging or capacitive)
or better biometrics?
or anything that uses CCD?

You have not described your objective.

edit
------

I just discovered Hitachi's vascular whitepaper.
http://www.hitachi.eu/veinid/documents/veinidwhitepaper.pdf

Ay biometric design must have a spec or goal.

  1. securty errorate ( false positive)
  2. recognition time ( capture,process, recognize)
  3. recognition error rate ( false negative)

vascular method is slow , complex, higher error rate but impossible to fake with latex.

  • so read whitepapers and more
  • then define goals, specs
 
Last edited:
Better what ?

scanning veins ? ( uses transmissive IR )
or finger scanner ? (uses refective imaging or capacitive)
or better biometrics?
or anything that uses CCD?

You have not described your objective.

edit
------

I just discovered Hitachi's vascular whitepaper.
http://www.hitachi.eu/veinid/documents/veinidwhitepaper.pdf

Ay biometric design must have a spec or goal.

  1. securty errorate ( false positive)
  2. recognition time ( capture,process, recognize)
  3. recognition error rate ( false negative)

vascular method is slow , complex, higher error rate but impossible to fake with latex.

  • so read whitepapers and more
  • then define goals, specs
Sir, I am trying to develop finger vein recognition device. before that i would like to take expert advise. Thanks for your help Sir. Now i planning to start with some IR diodes and CCD Camera.
 
Do you know How to process image takes significant matrix multiplication for edge filter, pattern recognition for fiducials or bifurcation points and converting 3D object by light transmission to 2D array requires , more matrix math to align image for orientation error??

Then image gets reduced to vectors and points, which contains signal and noise, each must be processed to identify what isnsignal,and what is noise, then statistically weighted according to SNR, then quantized, normallized for data base storage and parametric pattern recognition with tolerances.

The optical part may seem trivial but processing power cannot be done on a quad core snapdragon quickly.

Do you know all the design requirements before you start the design?
If determined to try in vein and you cannot perform task of defining all processes, as Hitachi says "is it in vain?"


Hitachi has great skill and they are still poor for recognition and slow, but at least "true negative " rate can be only 5% which may be disappointing even if done correctly while false positive can be avoided since it cannot be cloned with latex surface.

If any of this is foreign to you, then you are not ready to design, yet.

Start with FAR , FRR targets then available resources, then define inputs, data output and all data processes. with tolerances.
 
Last edited:
Do you know How to process image takes significant matrix multiplication for edge filter, pattern recognition for fiducials or bifurcation points and converting 3D object by light transmission to 2D array requires , more matrix math to align image for orientation error??

Then image gets reduced to vectors and points, which contains signal and noise, each must be processed to identify what is n signal,and what is noise, then statistically weighted according to SNR, then quantized, normallized for data base storage and parametric pattern recognition with tolerances.

The optical part may seem trivial but processing power cannot be done on a quad core snapdragon quickly.

Do you know all the design requirements before you start the design?
If determined to try in vein and you cannot perform task of defining all processes, as Hitachi says "is it in vain?"


Hitachi has great skill and they are still poor for recognition and slow, but at least "true negative " rate can be only 5% which may be disappointing even if done correctly while false positive can be avoided since it cannot be cloned with latex surface.

If any of this is foreign to you, then you are not ready to design, yet.

Start with FAR , FRR targets then available resources, then define inputs, data output and all data processes. with tolerances.


I planned some sketch outs. Here it is.
zvbgp4.png

Infrared is commonly divided into three regions that is near, mid and far-infrared light, but the boundaries between them are not agreed upon. There are two choices that focuses on imaging of vein patterns in hand by infrared light, the far-infrared (FIR) imaging and the near-infrared (NIR) imaging, which are suitable to capture human bodies images in a non-harmful way
29fat7t.jpg


Algorithm for vein pattern extraction consist of four steps:
  1. Image pre-processing
  2. Extraction of center positions of veins
  3. Connection of veins centers
  4. Binarization
bdkvnb.jpg

I started this implementation in opencv and c++
 
Top