Localized Dictionaries Based Orientation Field Estimation for Latent Fingerprints

Area of Science:

• Biometrics
• Forensic Science
• Computer Vision

Background:

• Dictionary-based methods offer good performance for latent fingerprint orientation field estimation.
• Existing methods can be further improved by incorporating stronger prior knowledge of fingerprint characteristics.
• Orientation patterns vary across different locations within a fingerprint.

Purpose of the Study:

• To develop a localized dictionaries-based algorithm for enhanced latent fingerprint orientation field estimation.
• To improve the accuracy and robustness of latent fingerprint analysis.
• To introduce a novel fingerprint pose estimation algorithm to support localized dictionary application.

Main Methods:

• A localized dictionaries-based orientation field estimation algorithm was developed.
• Noisy orientation patches were replaced with corresponding real orientation patches from local dictionaries.
• A Hough transform-based algorithm was proposed for latent fingerprint pose estimation.
• The pose estimation accumulates predictions from all orientation patches.

Main Results:

• The proposed localized dictionary approach significantly outperforms previous methods on challenging latent fingerprint datasets.
• The integrated Hough transform-based pose estimation enables effective application of localized dictionaries.
• Experimental results demonstrate marked improvements in orientation field estimation accuracy.

Conclusions:

• Localized dictionaries provide stronger prior knowledge for improved latent fingerprint orientation field estimation.
• The proposed method offers a significant advancement in latent fingerprint analysis and recognition.
• The Hough transform-based pose estimation is crucial for the success of the localized dictionary approach.