Robust fingerprint reconstruction using attention mechanism based autoencoders and multi-kernel autoencoders.

Fingerprint recognition technology is widely employed for identity verification and access control across diverse domains in which the quality of fingerprint images is critical for accurate biometric identification. However, fingerprint images can be damaged or incomplete due to various factors such as sensor limitations, environmental conditions, or physical injuries. To address this, several approaches including attention mechanism based autoencoders, multi-kernel sequential and multi-kernel stacked autoencoders, and multi-kernel ensemble autoencoders are proposed to perform reconstruction of incomplete and damaged fingerprints. Attention mechanisms play a crucial role by selectively emphasizing important regions and details, further enhancing information capture. By utilizing different autoencoder models with varying kernel sizes, local details and global context can be effectively captured, resulting in more precise restoration. The adoption of multi-kernel sequential and multi-kernel stacked autoencoders enables the extraction of increasingly abstract features, enhancing the model's ability to capture complex fingerprint characteristics. Utilizing autoencoder-based methods for the reconstruction of damaged or incomplete fingerprint images can enhance the accuracy and reliability of fingerprint identification systems leading to improved security and efficiency in real-world applications. Evaluation is conducted using image quality assessment metrics and feature-based matching is utilized to compute fingerprint matching accuracy. The multi-kernel ensemble autoencoder model produces the best reconstruction output with an average fingerprint matching accuracy of 93.81 %. The findings highlight the effectiveness of the proposed work in achieving high-quality reconstruction output and accurate fingerprint matching. The proposed work can be applied in various domains that include law enforcement, security, forensic analysis, and biometric authentication. [ABSTRACT FROM AUTHOR]