Overlapping Fingerprint Segmentation and Separation Based on Deep Learning

This study focuses on the application of semantic segmentation to address the challenge of overlapping fingerprint images with noisy backgrounds. Our approach involves excluding non-fingerprint regions and employing an innovative algorithm for separating and merging the overlapping regions with the non-overlapping areas of two fingerprints. This process allows us to obtain two distinct and independent fingerprint images. To mimic real-world scenarios encountered at crime scenes, we utilize an envelope bag image as the noisy background. We also incorporate artificial fingerprint images generated at various angles, adjust the distance between the two fingerprints, and superimpose them to simulate overlapping. These measures ensure a comprehensive evaluation of our method’s effectiveness. Through extensive experimentation, we report impressive results. The fingerprint segmentation model achieves an outstanding accuracy of 0.9894 on the test set. This accuracy confirms the model’s ability to robustly segment overlapping fingerprints. Our study contributes to advancing the field of fingerprint analysis by addressing the complex issue of overlapping fingerprints and providing a reliable solution. The achieved accuracy demonstrates the potential of our methodology in practical forensic applications, helping forensic personnel experts extract valuable information from challenging fingerprint datasets.