Slikovna biometrija v forenziki / Image-Based Biometrics in Forensic Science

Predavatelj bo predstavil težave, s katerimi se je srečal pri svojem delu kot sodni izvedenec za področje biometričnih prepoznav obrazov in interpretacije slik. Preden lahko na sliki iz nadzorne kamere uporabimo sistem za samodejno prepoznavo obrazov, jo moramo z različnimi metodami za obdelavo slik izboljšati in popraviti. Z metodo računalniškega vida iz slik rekonstruiramo 3D informacije. Včasih je treba potrditi tudi pristnost videomateriala, za kar lahko uporabimo različne tehnike na podlagi poznavanja formata zapisa in delovanja algoritmov za obdelavo slike in videa.

Če identifikacija na osnovi obraza ni mogoča ali je neuspešna, lahko uporabimo tudi druge biometrične karakteristike osebe − mehke biometrike. Sodni izvedenec za slikovno biometrijo mora razumeti in znati uporabiti različne metode za obdelavo slik in različna orodja računalniškega vida. Sposoben mora biti tudi razložiti svoje ugotovitve in koliko verjetni so končni rezultati.

---

The lecturer will present the problems he encountered in biometric face recognition and biometric image interpretation in their experience as court appointed expert witnesses. Before an automated face recognition system can be applied on a typical surveillance video, images must be enhanced using various image-processing methods or enriched by using computer vision 3D reconstruction methods. Authenticity of video material must also sometimes be verified, for which we can use various techniques based on the knowledge of the format and the operation of image and video processing algorithms.

If face recognition is not possible or successful then other soft biometric characteristics can be checked. A legal expert witness for image biometry must be able to employ a large array of image processing and computer vision tools and methods. The expert witness must be able to explain how the biometric results were obtained, what the necessary processing steps were, and how confident the final results are.