Tiny Partial Fingerprint Sensor Quality Assessment

Tiny partial fingerprint sensor image recognition is pivotal for busi-ness and security, increasingly integrated with machine and deep learning. Image quality is crucial, with NIST's NFIQ2, OpenAFQA, and MiDeCon leading. Current methods focus on larger full fingerprint sensor datasets, incurring high costs and hindering retraining on alternative data. This poses significant challenges, particularly for Artificial Intelligence of Things (AIoT) edge devices and the analysis of tiny partial fingerprints. To overcome these challenges, we propose a novel metric called the ridge spectrum indicator. This metric is uniquely designed to cater to large-area and tiny partial fingerprint datasets, demonstrating superior performance when compared with existing methods. We present an automatic labeling system seamlessly integrated with the K nearest neighbor (KNN) Classifier, eliminating the need for dependency on other assessment methods, and thereby streamlining the retraining process. Our approach achieves 98% accuracy in assessing partial fingerprint quality, overcoming current limitations, and enhancing efficiency in fingerprint recognition, especially for AIoT edge devices and tiny partial fingerprint sensor datasets. This novel solution's 176 × 36 pixel dimensions save at least 1/10 of the sensor area wafer cost.