Computer Vision Projects
- IRIS based identification
I am working on IRIS of a human eye based identification problem, which is a part of my Master thesis.
Steps to person recogniting :
System has first to determine the distance between the object and the camera.This step is important because of two reasons. Object should not be too far of camera (it has to be in range of 5cm to 40cm) and if the sistem does not contain autofocus and to decide if picture is focused or it is not. I didn't use camera autofocusing and auto tracking system it has to be done with some software procedures. I have implemented autofocus algorithm based on absolute diference between two mediana filter values.
Captured and focused color or BW picture of the eye
Because of time comsuptioning algorithms I had to decrease image resolution and first applay my circle searching algorithm on pictures with low resolution to get approximate results.
250 x 250
170 x 170
56 x 56
Sobel edge detector operator is used to precisely define the edge position. Picture is then binarised.
Circle finding algorithm is applied to find an outer circle (IRIS) and an inner circle (pupil of the eye).
This picture is then transformed to some easier form to get a binary code from it.
Picture is then convolved with a filter mask (several diferent tipes of filters were tested) and the result is a binary code, which can be compared to some codes from database.
At the moment I am testing also a new aproach and some new methods for solving the same problem.
I have also made some hardware stuff for our LAB needs
- High-eficient circular LED lights for Computer Vision purpose
M. Knez, S. Kovacic, Dolocanje polozaja oci s pomocjo deformacijskih modelov
(Determing position of the eyes based on deformable models matching)