[Paper Review] DWT Based Fingerprint Recognition using Non Minutiae Features

Deepak Kumar, K B Raja+2 more|arXiv (Cornell University)|Jun 17, 2011

Biometric Identification and Security25 references12 citations

TL;DR

This paper proposes a DWT-based fingerprint recognition system using non-minutiae features to improve accuracy in forensic and security applications. By applying 3-level discrete wavelet transform (DWT) to decompose fingerprint images, extracting dominant orientation and coherence from the LL band, and computing center area and edge features across all subbands, the method uses Euclidean distance for matching—achieving lower FAR, FRR, and TSR than existing approaches.

ABSTRACT

Forensic applications like criminal investigations, terrorist identification and National security issues require a strong fingerprint data base and efficient identification system. In this paper we propose DWT based Fingerprint Recognition using Non Minutiae (DWTFR) algorithm. Fingerprint image is decomposed into multi resolution sub bands of LL, LH, HL and HH by applying 3 level DWT. The Dominant local orientation angle θ and Coherence are computed on LL band only. The Centre Area Features and Edge Parameters are determined on each DWT level by considering all four sub bands. The comparison of test fingerprint with database fingerprint is decided based on the Euclidean Distance of all the features. It is observed that the values of FAR, FRR and TSR are improved compared to the existing algorithm.

Motivation & Objective

• To develop a robust fingerprint recognition system for forensic and national security applications.
• To reduce dependency on traditional minutiae-based features, which are sensitive to image degradation.
• To improve recognition accuracy by leveraging non-minutiae features derived from multi-resolution DWT decomposition.
• To minimize false acceptance rate (FAR), false rejection rate (FRR), and total success rate (TSR) through feature-based matching.
• To evaluate performance against existing algorithms using standard metrics in fingerprint recognition.

Proposed method

• Applying a 3-level Discrete Wavelet Transform (DWT) to decompose the fingerprint image into subbands: LL, LH, HL, and HH.
• Computing the dominant local orientation angle (θ) and coherence from the LL subband only, which captures the main structural information.
• Extracting Center Area Features and Edge Parameters from all four DWT subbands at each decomposition level.
• Using Euclidean distance to compare test fingerprint features with those in the database for matching.
• Aggregating features across multiple DWT levels to enhance discriminative power and robustness.
• Employing a feature vector combining orientation, coherence, center area, and edge parameters for classification.

Experimental results

Research questions

• RQ1 Can non-minutiae features derived from DWT improve fingerprint recognition accuracy compared to traditional minutiae-based methods?
• RQ2 How does multi-resolution DWT decomposition enhance feature extraction in noisy or degraded fingerprint images?
• RQ3 To what extent do the extracted features (orientation, coherence, center area, edge parameters) contribute to reducing FAR and FRR?
• RQ4 How does the proposed DWT-based matching strategy compare in performance to existing fingerprint recognition algorithms?
• RQ5 Can the integration of features across multiple DWT subbands improve recognition reliability and robustness?

Key findings

• The proposed DWTFR algorithm achieves improved performance in terms of lower false acceptance rate (FAR) compared to existing methods.
• The false rejection rate (FRR) is reduced, indicating better reliability in matching genuine fingerprint samples.
• The total success rate (TSR) is enhanced, demonstrating overall higher accuracy in fingerprint recognition.
• The use of non-minutiae features such as orientation, coherence, and edge parameters contributes significantly to robustness against image degradation.
• Feature extraction from the LL band for orientation and coherence, combined with multi-level DWT subband analysis, enhances discriminative capability.
• The Euclidean distance-based matching strategy effectively compares feature vectors, leading to consistent and accurate identification outcomes.