| Current multimedia communication standards such as JPEG, MPEG, H.263, are extremely vulnerable to network errors (bit errors and packet losses). In this dissertation, we propose a unified framework for error resilient multimedia transmission at the application layer. In this framework, the responsibility of the end-to-end image/video quality is a cooperative effort between the sender and the receiver. Our unified framework provides the following advantages: (a) robust and efficient compression, (b) low FEC overhead, and (c) fast channel decoding. Efficient and robust compression is achieved by using a new pixel-level interleaving scheme in the transform domain, powered up with a new concept of Variable End-of-Block (VAC-EOB). Information inside the block is decorrelated, so that network errors on the transmitted image/video information are spread out over larger and uncorrelated regions. This increases the chances of a good reconstruction process at the receiver. Low overhead protection and fast channel decoding is achieved using a Variable and Selective FEC (VS-FEC) scheme. In this scheme, the amount of data protection depends on the current network error rate provided by the receiver. Additionally, our scheme is selective in the sense that only important information is protected. This is different from the current FEC schemes, where the entire information is protected (increasing the bandwidth requirements). The amount of FEC needed for protection is based on a 4-state Markov Model, which predicts the probability of losing the important information.; In the case of image transmission, if the received information still has errors, we propose a new geometric EC scheme to eliminate or comprehend with the errors. The proposed scheme extracts useful information, such as edge curvature and average tangent from healthy blocks around the damaged block of the received image. This information is input into a Bezier polynomial, which approximates shape of the lost edges for their reconstruction. We show that our proposed scheme performs 2%–11% much better than current compression standards, and produces a total overhead protection of 5.5 times better than Channel/coding Error Control schemes (or linear Error Control Schemes). |
