3D non-contact / touchless Fingerprints

Research--3D Non-contact / Touchless Fingerprint

3D Non-contact / Touchless Fingerprints

[1] Data Acquisition and Processing of 3-D Fingerprints.

[2] Quality and Matching Performance Analysis of Three-dimensional Unraveled Fingerprints.

[3] Fit-sphere Unwrapping and Performance Analysis of 3D Fingerprints.

[4] Data Acquisition and Quality Analysis of 3-Dimensional Fingerprints.

System Overview

Fingerprint recognition has been extensively applied in both forensic law enforcement and security applications involving personal identification. However, limitations are imposed upon the current fingerprint capture technologies including:
1) obligatory maintenance of a clean sensor or prism surface;
2) uncontrollability and non-uniformity of the finger pressure on the device;
3) permanent or semi-permanent change of the finger ridge structure due to injuries or heavy manual labors;
4) residues from the previous fingerprint capture;
5) data distortion under different illumination, environmental, and finger skin conditions; and
6) extra scanning time and motion artifacts incurred in technologies that require finger rolling.

In order to overcome these limitations, we have been developing a contactless / touchless 3D scanning system that employs structured light illumination (SLI). Our ultimate goal is to simultaneously acquire 3D scans of all the five fingers and the palm in high speed and fidelity using multiple, commodity digital cameras and a DLP projector. Post processing of these scans is then performed to virtually extract the finger and palm surfaces, and create 2D flat equivalent images.

The advantages of our 3D fingerprint scanning and processing technology include:
1) non-contact defuses distortion that exists in conventional fingerprint acquisition system.
2) simultaneous acquisition of both texture and ridge depth information of fingers;
3) automated fingerprint entry in no need of interaction with the operator;
4) fast scanning (less than 0.8 second);
5) robustness to contamination of fingers and residues of previous users;
6) robustness to clutter and fraud because of the difficulties in faking 3D fingerprints;
7) real-time feedback (less than 5 seconds) for users to make position adjustment; and
8) low cost by using the off-the-shelf commodity camera and projector.

* The 3D contactless / touchless fingerprint scanner. The size of acquired image is 1420*2064 such that the ppi value can be as high as 1500.

* A 3D noncontact / touchless fingerprint scan. From left to right are: a 3D finger with texture (surface reflectance variations or albedo image); 3D fingerprint with depth rendering; cropped 3D fingerprint; and rotated crop.

3D Data Processing

To be compatible with current 2D automatic fingerprint identification system (AFIS), we developed a new flattening algorithm for the 3D noncontact / touchless fingerprint which obtains both depth and albedo flattened fingerprints. The two fingerprints are further fused to obtain the final flatten result. After 3D fingerprint scans are unraveled into 2D flat equivalent fingerprints. The 2D flat equivalent fingerprints are further processed to be compliant with the National Institute of Standard and Technology (NIST) standards.

* Overlapping between 2D ink rolled fingerprints and flattened 3D fingerprints. 2D ink rolled fingerprints are binarized and in the left parts, whereas the corresponding flattened 3D fingerprints are in the right parts. (a) Three combined images from 2D ink rolled fingerprint and depth flattened 3D fingerprint. (b) Three combined images from 2D ink rolled fingerprint and albedo flattened 3D fingerprint.

* The fingerprint scan. From left to right are: a 2D plain fingerprint; the corresponding 3D scan; the 2D cropped unraveled equivalent fingerprint obtained from 3D scan.

Performance

The matching performances of depth, albedo and fused fingerprint data sets are studied, using BOZORTH3 by NIST software package. BOZORTH3 employs features to minutiae of the fingerprints and produces a real valued similarity score. The higher the score is, the more likelihood that the two fingers are from the same finger of the same subject. We note the output value as genuine score if the input two fingerprints are actually from the same finger of the same subject, whereas impostor score if the two fingerprints are from the same finger but of different subjects.

* Distribution of genuine and impostor scores for depth, albedo and fused fingerprints. Result is obtained from 1011 genuine scores and 1011 impostor scores.

Based on the distributions, we give the ROC which is a statement of the performance of the fingerprint verification system. To evaluate the performance, we have to set the operating threshold. For each threshold, we compute out the False Accept Rate (FAR) and True Accept Rate (TAR). For a generally specific FAR 0.01, the TAR of the fused data set is 0.98, while the TAR of the depth and albedo are 0.93 and 0.96. All the three data sets achieve high matching performance, and the fused set outperforms the depth and albedo sets in ROC.

* ROC of depth, albedo and fused flattened fingerprints.

For more details, please visit Flashscan3D,LLC or go to Yongchang's home.