| The recent advent of wireless devices capable of communicating with images has made possible a radical new approach to mobile communications. For the first time one can access, from almost anywhere, information previously available only at fixed locations. The bandwidth and storage space available to these devices, however, will likely remain sharply restricted in the near term. For their users, image distortion due to lossy compression will be a fact of life. This does not mean that images need be undecipherable, rendered useless by compression artifacts.; This dissertation is directed towards lossy compression of images for low bandwidth, low bit rate environments, with a view to maintaining users' ability to recognize and usefully interpret the compressed images. It contributes novel approaches in two categories: progressive compression algorithms, and experimental evaluation of these algorithms.; In the first category, we present several new coders designed to improve viewers' ability to understand the content of highly compressed images. The first coder is a variation on SPIHT that presents images in a stepwise progression of increasing resolution. It transmits only information relevant to the current resolution, deferring other information until later in the bitstream. We also present several coders for images containing combinations of text, graphics and photographs. These coders localize text in the images, and improve its visual quality by using specialized filtering and other techniques. Finally, we introduce two coders that preserve the clarity of sharp-edged image features at low bit rates. We present results indicating that each of the above coders improves visual performance over standard coders within its particular area of applicability.; In the second category, we present a novel experimental and statistical framework for comparing progressive coders. The comparisons use response time studies in which human observers view a series of progressive transmissions, and respond to questions about the images as they become recognizable. We use the framework to compare several well known algorithms (JPEG, EZW, SPIHT), and to show that a multi-resolution decoding is recognized faster than a single large scale decoding. Our experiments also show that global blurriness slows down recognition more than do localized “splotch” artifacts. |
