Preprocessing Latent-Fingerprint Images For Improving Segmentation Using Morphological Snakes

Latent fingerprints have played a critical role in identifying criminals and suspects. However, latent fingerprint identification is more complicated than plain and rolled fingerprints mainly due to poor ridge quality, complex background noise and overlapped structured noise in latent images. Subsequently, a latent-fingerprint image requires to be segmented to extract the fingerprint region from the background. The paper proposes a novel and efficient technique for latent-fingerprint segmentation. Our approach is based mainly on two fundamental ideas: i) applying the conversion from RGB color model to YCBCR color model and the Gaussian blur technique as a preprocessing before segmentation, and ii) using morphologic active contours without edges to define the fingerprint region based on an evolving contour that starts its rapid evolution in a stable state from the inside fingerprint. The technique is tested on two fingerprint databases: FVC2004 and NIST SD27. Our experimental results evaluate the miss-classified pixels and yield high segmentation accuracy.