Fingerprint recognition stands as one of the oldest and most dependable biometric authentication techniques, widely employed in contexts from mobile device security to forensic investigations. Its robustness arises from the uniqueness and permanence of ridge patterns on human fingers, which provide a reliable means to identify individuals.

The technology faces two central challenges that affect its reliability and security. First, fingerprint systems are vulnerable to spoofing attacks, where attackers attempt to fool the system using artificial or fake fingerprints created from materials such as silicone, gelatin, or 2D prints. These sophisticated attacks pose significant security threats. Second, latent fingerprints, often recovered from crime scenes, suffer from poor quality due to noise, low clarity, overlapping prints, and environmental conditions, making accurate identification difficult.

In recent years, these challenges have been addressed through deep learning techniques, particularly with the advent of Convolutional Neural Networks (CNNs). Deep learning models learn hierarchical, detailed features directly from raw fingerprint images, empowering systems to more effectively distinguish subtle differences in fingerprint characteristics. CNN-based approaches excel at both spoof detection and latent print enhancement. For spoof detection, CNN architectures such as the VGG16 network have proven highly effective, known for its depth and effectiveness in feature extraction.

VGG16- It is a deep convolutional neural network architecture with 16 layers of weights, designed for image classification and recognition tasks. It consists of multiple layers of small 3x3 filters, organized into blocks, followed by fully connected layers, and is known for its simplicity and high accuracy.

CNN based fingerprint liveness detection models train on large datasets containing both real and spoof fingerprint images, learning subtle discriminative traits between genuine and fake fingerprints. Alongside classical image pre-processing steps like grayscale conversion, HSV color transformation, and edge detection (e.g., Canny), these models improve their ability to focus on crucial fingerprint features.

Latent fingerprint enhancement and segmentation involve deep learning networks designed to clarify and separate overlapping fingerprint patterns in noisy and degraded images. Mask R-CNN and other segmentation models implement advanced normalization and noise filtering to restore ridge clarity and isolate individual fingerprints. This automated enhancement greatly supports forensic analysts by enabling

more accurate minutiae extraction and matching in difficult cases.

Integrating CNN-based liveness detection, latent print enhancement, and expert verification, is a promising direction to boost fingerprint recognition reliability. This review paper examines recent works on using deep learning for (1) fingerprint liveness detection and (2) latent fingerprint enhancement and segmentation, highlighting their results and limitations.