Study On Multispectral Latent Fingerprint Enhancement Based On Frequency Domain Transform

In recent years, fingerprint recognition has made remarkable progress as the representation of biometric technology, and latent fingerprint left at crime scenes has played a prominent role in the field of criminal investigation as a special kind of vestigial evidence, so it has an extremely important research value and attracted highly attentions by scholars and Governments from various countries. Generally recognized as " The First Evidence " by the judiciary, crime scene latent fingerprints are the necessary condition for recognizing and identifying fingerprints correctly or even found the suspect eventually, but due to some unique characteristics of the latent fingerprints such as ambiguity, fragmentary nature, complex background, ridge line discontinuities, ridge overlapping, it is still in a bottleneck state so far to find an effective detecting and accurate certificating technology which restricted the development of scene latent fingerprints identification technology seriously while limited its further application and promotion.Mid-infrared analysis is based on the absorption differences of infrared light by materials with different chemical structure. Fingerprint residue contains esters, small proteins and hydroxyl acid on skins which characteristic peaks are located in mid-infrared region and can fully reflect the full spectrum image data of scene fingerprint information is formed by spectral information from a few hundreds to several thousands wave-number.In the process of exploring and researching methods and theories fit to deal with difficult scene latent fingerprint images, multi-spectral non-destructive analysis based on the SpotLight400 infrared imaging system is used to analyze fuzzy fingerprint traces for the first time. Multi-spectral image dimension is reduced by pre-processing algorithms which are combined with spectral subtraction and principal component analysis (PCA) and optimal spectral fingerprint images are obtained. The method that using overlapping neighbors domain image block to select minimum noise impact of neighborhood as the base leading direction is utilized in order to receive reliable orientation field, and all the subsequent frames after dimension reduction are enhanced applying Gabor filter based fingerprint enhancement algorithm, it is proved that the effects are pretty good for fingerprint images of large creases, indurations, wet or influential spots. At last, the multi-spectral fingerprint image fusion technique based on wavelet transform is applied to get high quality fingerprint images.In this paper, multi-spectral data dimension is reduced using principal component analysis and the frames after dimension reduction are enhanced utilizing fingerprint enhancement algorithm based on Gabor filter, finally, wavelet transform based fusion algorithm of multi-spectral fingerprints is conducted and the effect proved to be good. The experimental result shows that the feasibility, accuracy and effectiveness of this algorithm.