|
|
Authenticating Multimedia In The Presence Of Noise |
|
by Emin Martinian The authenticity of multimedia such as photos, videos, and sound recordings is critically important in many applications. Unfortunately, due to the power and economy of modern digital processing tools, creating convincing forgeries is relatively easy. Consequently there is a need for multimedia authentication methods to restore credibility. Various researchers have proposed techniques based on digital watermarking, cryptography, and content classification. However, since authenticating multimedia is a fairly new idea, measures for evaluating different methods are not well defined. After introducing some possible applications to clarify the problem and discussing some previously proposed methods, we present a framework for authenticating multimedia under the title `Analog Authentication'. Defining a precise problem including the performance measures, tradeoffs, and resources available provides a way to evaluate different schemes. In addition, we analyze the fundamental limits of what can and can not be done by any scheme. Our analysis yields a quantity called the fundamental distortion which serves as the dividing line between what is asymptotically achievable and what is impossible. This is analogous to how the capacity of a communications channel characterizes which transmission rates are achievable and which are impossible. In addition our analysis is constructive and yields insight into practical methods. Finally, we design some practical analog authentication schemes and evaluate their performance. Download Thesis |
|