Progressive Latent Fingerprint Enhancement Using Two-Stage Spectrum Boosting with Matched Filter and Sparse Autoencoder

We propose a progressive enhancement algorithm to improve friction ridges in latent fingerprint images that are usually of poor quality or partially missing. The proposed method is designed for performing in the frequency domain to propagate a friction ridge pattern from good to poor-quality areas. In the first stage, the algorithm starts with a block of high ridge signal strength to initially boost ridge spectra by a matched filter. The boosted block is then padded back to the input image, so that neighboring blocks can absorb the stronger ridge signal. We carry on this process iteratively at neighbors to propagate the friction ridge pattern until the entire image is done. However, at the low ridge signal strength, the matched filter cannot enhance the ridge signal well enough. In the second stage, we use a sparse autoencoder-based spectrum generator to approximate the matched filter for the spectrum boosting process. The proposed method was benchmarked with two existing latent fingerprint enhancement methods. The experimental result shows that the proposed method provided the promising accuracy on publicly available IIT-D MOLF latent fingerprint database.