Although low-quality fingerprints have been approved and verified successfully in fingerprint verification competitions (FVC), the efforts to determine methods to improve fingerprint quality are ongoing. Additionally, low-computation-complexity enhancement algorithms are required for actual industrial applications. The main core operations are performed and easy parallel implementation is achieved owing to the autoencoder’s architecture, which is composed of convolutional and polling layers. This paper presents a low-cost enhancement method to improve low-quality fingerprint images based on the autoencoder’s architecture. The experimental results indicate that the enhancement of the image quality score is higher when our proposed method is used, yielding an approximate average improvement of 2 units for images stored in the public databases (DBs) FVC2000 DB2, FVC2002 DB3, and FVC2004 DB2.
Hsu et al. (Mon,) studied this question.