A new system for securizing documents through their entire life cycle

The authenticity of a printed document is taken for granted upon the existence of a handwritten signature or some sort of stamp. The most widespread method used to ensure that a printed document is not tampered with is the mere visual inspection of the document to be verified and of a trusted copy. This is a highly inefficient process. Frauds involving paper documents are very easy to be realized and very difficult to be detected. Only in the digital world, thanks to the “digital signature” (i.e., the encrypted hash of the file to be signed), documents are fully protected from the threat of unauthorized alteration. Unfortunately, during their lifetime documents may be repeatedly printed and scanned, and digital to paper conversions result in loosing the digital signature. Documents are unprotected from the first time they leave the digital world. The major contribution of this thesis is the definition, design and development of a new system for securizing documents through their entire life cycle. Another key contribution is the design of a novel high capacity barcode, denoted as HCC2D code, without which it would have been hard to embed all the information needed for verifying a document within the document itself: HCC2D codes achieve an effective data density of 3,972 bytes=inch2 with a success rate of 90%, while data densities of stateof- art barcodes are under 2; 000 bytes=inch2. Our system reaches the following goals: to embed all the security information within the paper itself; to be applicable to any type of document; to identity the regions of the document that have been tampered with; to support identity verification of the user presenting the document; to verify the authenticity of the handwritten signature. To the best of our knowledge, there is no existing solution which is able to address all of those critical points simultaneously. The experimental evaluation of our system gives interesting results. It gets a negligible probability of not detecting a fraudulent manipulation (under 1%) and of gener-ating a false alarm (under 1%). In particular, a malicious manipulation is detected in 99% of cases. As for face recognition and handwritten signature verification (HSV), we get good results, despite storage constraints due to 2D barcodes: precision and recall curves cross at around 90% for face recognition and at around 89% for HSV.