Convolution Comparison Pattern: An Efficient Local Image Descriptor for Fingerprint Liveness Detection.

Area of Science:

• Biometrics
• Computer Vision
• Image Processing

Background:

• Local image descriptors are crucial for pattern recognition tasks.
• Existing descriptors face challenges in rotation invariance and robustness for specific applications like liveness detection.

Purpose of the Study:

• To introduce a novel local image descriptor, Convolution Comparison Pattern (CCP).
• To evaluate CCP's effectiveness for fingerprint liveness detection.
• To explore CCP's potential in other image analysis domains.

Main Methods:

• Generating rotation-invariant image patches using fingerprint segmentation and orientation fields.
• Computing Discrete Cosine Transform (DCT) on patches.
• Creating binary patterns by comparing DCT coefficients.
• Summarizing patterns into histograms and concatenating them into feature vectors for classification.

Main Results:

• The proposed CCP descriptor demonstrates significant utility in fingerprint liveness detection.
• CCP outperforms established local descriptors (LBP, LPQ, WLD) on the LivDet 2013 benchmark.
• Experimental results validate the effectiveness of the CCP descriptor.

Conclusions:

• CCP is a novel and effective local image descriptor.
• CCP shows superior performance in fingerprint liveness detection compared to existing methods.
• CCP holds promise for diverse applications including biological imaging, texture, face, and iris recognition, and machine vision.