Authenticated Image Compression Abstract

Digital images are valuable commodity in today's society. As a result, the ability to store, access, or transmit them in an efficient and secure manner has become crucial. Since a large number of bits is typically required to represent even a single image, many good image compression schemes have been developed over the last few decades. These schemes led to various efficient representations of images.

As more and more digital images are distributed via the internet, or stored in databases, the issue of protecting these images from fraud attacks becomes a necessity. Embedding a digital watermark (also known as digital signature) in a given image is an authentication scheme which is capable of detecting any change made to an image, including changes in pixel values and image size, without requiring the existence of the original image. However, non of the existing authentication schemes attempt to predict the source of these changes, i.e., whether these changes are due to a fraud attack, transmission errors, filtering, or lossy compression, although each of these sources has different nature. Moreover, if a digital watermark is embedded first, followed by applying lossy compression, the authentication scheme may fail to validate the image, due to the loss of information during compression. On the other hand, if the compression scheme is applied first, the digital watermarking scheme may not be able to benefit from the two-dimensional nature of the image data, due to the higher and complex representation of the compressed data. Also, most likely the decompression module will not be able to produce the image, since decompression is sensitive to any altering in the decompressed data.

This proposed research will attempt at developing a new class of compression schemes with authentication capability. This class of authenticated image compression schemes allows the watermarking process to benefit from the two-dimensional nature of the image data. Moreover, it protects the authentication process from failure, due to the loss of information during compression. Furthermore, it facilitates a progressive authentication, by which a partial transmission of an image can still be authenticated. This novel feature is extremely important for several imaging applications, including trusted camera, commercial image transaction, legal usage of images, the archival and retrieval of medical images. This feature is not available in any of the existing authentication systems. Finally, this research will also investigate and develop authentication schemes that can predict the source of the change, within a certain confidence.