Denoising of fingerprint images by exploring external and internal correlations

Fingerprint is an important measure used to detect an unknown victim, suspect or witness. It has a major role in verifying records to explore links and matches between a suspect and a crime. Fingerprints are also used for security reasons, such as an entrance control at important buildings. But the quality of fingerprint images can easily get degraded by skin dryness, wet, wound and other types of noises. Hence denoising of fingerprint images is a necessary step in systems for automatic fingerprint recognition. This paper suggests a 3-stage process for the removal of noise from fingerprint images, through exploring external correlations and internal correlations, with the help of a set of correlated images. Internal and external data cubes are built for each noisy patch by discovering identical patches from the corresponding noisy and internet based images. External denoising in the first stage is done by a graph based optimization method and internal denoising is done by means of a frequency truncation process. Internal denoising results and external denoising results are combined to obtain the preliminary denoising result. The second stage performs filtering of external and internal cubes and the fused result is in turn passed to the third stage. In the third stage, an image enhancement technique is carried out to obtain the final denoised result. This method is compared with the existing algorithms and the experimental results, in terms of its PSNR (Peak Signal to Noise Ratio) values and SSIM (Structural Similarity Measure) values proved that the method is efficient than all of them.