Correntropy-Induced Discriminative Nonnegative Sparse Coding for Robust Palmprint Recognition

Kunlei Jing, Xinman Zhang, Guokun SongSensors 2020 · OpenReview

TL;DR

This paper introduces a novel method called Correntropy-Induced Discriminative Nonnegative Sparse Coding (CDNSC) for robust palmprint recognition, which effectively handles various image corruptions like occlusion and noise. The method combines a sophisticated error estimation technique with a discriminative nonnegative sparse regularizer, leading to significantly improved recognition accuracy, especially in challenging real-world scenarios.

What Problem Does This Paper Address?

Palmprint recognition is a promising biometric technology, but its real-world application is hindered by image degradations such as occlusions, corruptions, and variations in illumination and pose. Existing methods often struggle with these challenges, leading to reduced accuracy. This paper addresses the critical need for robust palmprint recognition systems that can reliably identify individuals even when their palmprint images are contaminated or incomplete. The authors specifically highlight the lack of in-depth investigations into robust palmprint recognition and aim to fill this gap by developing a method that can effectively detect and correct errors in palmprint images.

Methods

The core of this work is the Correntropy-Induced Discriminative Nonnegative Sparse Coding (CDNSC) method. It's built upon two key components:

  • Cooperative Error Estimator (CEE): This component combines the correntropy metric for robust error detection (especially against non-Gaussian noise and outliers) with the l1-norm for error correction. This synergistic approach allows for flexible and robust handling of various contaminations.
  • Discriminative Nonnegative Sparse Regularizer (DNSR): This novel regularizer is designed to extract significant, nonnegative features. It encourages sparsity in the representation coefficients and promotes discriminability between different palmprint classes.

The authors also propose an efficient optimization approach to handle the challenging non-negative constraint, converting it into an equality constraint. For multispectral palmprint recognition, they extend CDNSC to E-CDNSC by developing a constrained particle swarm optimizer (CPSO) to learn optimal fusion parameters for combining features from different spectral bands.

Datasets

The authors evaluate their methods on two publicly available multispectral palmprint databases:

  • CASIA database: A contactless palmprint database with variations in illumination and pose. It includes images from six spectral bands (460 nm, 630 nm, 700 nm, 850 nm, 940 nm, and white).
  • PolyU database: A contact-based palmprint database with restricted hand posture. It includes images from four spectral bands (red, blue, green, and NIR).

Key Insights and Ideas

  • Correntropy for Robustness: The correntropy metric is a powerful tool for error detection due to its robustness against non-Gaussian noise and outliers, making it ideal for real-world contaminated images.
  • Synergy of Error Detection and Correction: Combining a robust error detector (correntropy) with an error corrector (l1-norm) in a cooperative framework significantly enhances the system's ability to handle complex contaminations.
  • Importance of Discriminative Nonnegative Features: Encouraging nonnegative and discriminative sparse representations is crucial for extracting meaningful features that improve classification accuracy, especially when dealing with similar palmprints from different classes.
  • Multispectral Fusion for Enhanced Robustness: Extending the method to multispectral data and using CPSO for learned fusion parameters can further boost recognition performance and robustness by leveraging complementary information from different spectral bands.
  • Analytical Solutions are Key: Developing efficient analytical optimization methods, particularly for the non-negative constraint, is vital for practical implementation and computational efficiency.

Key Results

  • Superior Robustness: CDNSC consistently outperforms state-of-the-art methods in various challenging scenarios, including dense corruption, scar occlusion, and mixed contaminations, across both contactless (CASIA) and contact-based (PolyU) databases.
  • High Recognition Rates: For instance, on the CASIA database under 50% corruption, CDNSC achieved a recognition rate of 94.75%, significantly higher than other methods. In multispectral recognition on the CASIA database, E-CDNSC achieved a recognition rate of 98.5% in pure conditions and 97.75% under mixed contaminations.
  • Efficiency: While some traditional methods are faster, CDNSC offers a better tradeoff between accuracy and efficiency compared to other advanced robust methods.
  • Effective Fusion: The learned fusion parameters for multispectral recognition proved effective, leading to improved performance even when applied to contaminated samples.

Limitations and Future Work

  • Computational Complexity: While CDNSC is relatively efficient, the optimization process, especially for the matrix inversion, can be time-consuming compared to simpler methods.
  • Generalizability: The paper focuses on palmprint recognition. The authors suggest that the active ideas in CDNSC and E-CDNSC could be applied to other single-mode biometric recognition or multimodal biometric recognition systems.
  • Parameter Tuning: Like most machine learning methods, CDNSC and its extensions involve several parameters that might require careful tuning for optimal performance in different scenarios.

The paper's contributions lie in its novel combination of correntropy for error estimation and a discriminative nonnegative sparse regularizer, leading to a highly robust palmprint recognition system. The extension to multispectral recognition further enhances its capabilities.

Summary updated

Abstract

Palmprint recognition has been widely studied for security applications. However, there is a lack of in-depth investigations on robust palmprint recognition. Regression analysis being intuitively interpretable on robustness design inspires us to propose a correntropy-induced discriminative nonnegative sparse coding method for robust palmprint recognition. Specifically, we combine the correntropy metric and l1-norm to present a powerful error estimator that gains flexibility and robustness to various contaminations by cooperatively detecting and correcting errors. Furthermore, we equip the error estimator with a tailored discriminative nonnegative sparse regularizer to extract significant nonnegative features. We manage to explore an analytical optimization approach regarding this unified scheme and figure out a novel efficient method to address the challenging non-negative constraint. Finally, the proposed coding method is extended for robust multispectral palmprint recognition. Namely, we develop a constrained particle swarm optimizer to search for the feasible parameters to fuse the extracted robust features of different spectrums. Extensive experimental results on both contactless and contact-based multispectral palmprint databases verify the flexibility and robustness of our methods.
BibTeX
@inproceedings{jing2020correntropy,
  author = {Kunlei Jing and Xinman Zhang and Guokun Song},
  title = {{Correntropy-Induced Discriminative Nonnegative Sparse Coding for Robust Palmprint Recognition}},
  booktitle = {Sensors},
  year = {2020},
  url = {https://openreview.net/forum?id=DL1RnCtz9at},
}
Figure