Fingerprint Identification With Shallow Multifeature View Classifier.

Area of Science:

• Biometrics
• Computer Vision
• Machine Learning

Background:

• Deep Convolutional Neural Networks (DCNNs) are state-of-the-art for fingerprint classification but require extensive annotated data.
• Deeper DCNN architectures learn more abstract features, improving accuracy but increasing data demands.

Purpose of the Study:

• To propose an efficient fingerprint classification system that reduces search-space.
• To introduce a Shallow Multifeature View CNN (SMV-CNN) classifier that is less data-intensive than deep DCNNs.

Main Methods:

• The SMV-CNN extracts both fine-grained and abstract features from multiple representations of the input fingerprint image.
• These multi-view features are processed by a fully connected neural network for global classification prediction.
• The system integrates classification with minutiae neighbor-based feature encoding and matching for identification.

Main Results:

• SMV-CNN achieved a 2.8% improvement over a baseline CNN on an open-source database.
• The proposed method performs comparably to ResNet-50 but is less complex and more efficient for training.
• Fingerprint identification accuracy was maintained while reducing the search space by over 50%.

Conclusions:

• The SMV-CNN offers an efficient and effective approach to fingerprint classification, outperforming baseline CNNs.
• This method provides a viable alternative to deep DCNNs, requiring less data and computational resources.
• The classification-driven search-space reduction significantly enhances the efficiency of fingerprint identification systems.