||Delivering visual contents over heterogeneous networks (such as the Internet) is becoming more and more popular in recent years, partially due to the extraordinary image/video presentation capability and partially due to the increasing deployment of broadband network services. However, the varying channel capacity and unreliable transmission in heterogeneous networks make it quite challenging to provide the quality of service (QoS) to end-users, especially for delivering visual contents. One solution to handle the varying channel capacity is to use a scalable source coding scheme that can delicately adapt to available bandwidth. However, scalable coding solves only part of the problem, as QoS still cannot be guaranteed due to the unreliable transmission in heterogeneous networks. To combat with packet losses and bit errors coming from communication channels, many efforts have been devoted at different stages of the delivering process. However, traditional schemes are not able to guarantee QoS under practical time-varying channel conditions, because they are designed for specific environments under specific assumptions. Moreover, as many unique features originated from the scalable coding are not considered fully, these schemes are also lack of error resilience. In this thesis we produce a robust image/video delivering system that can deal with channel errors much more effectively. We define an metric to measure the delivering quality quantitatively. Based on the performance metric, we break down the complicated delivering system into four key functional modules, namely, bitstream packetization, packet loss protection, error detection, and error concealment, and provide optimization method for each of them. Firstly, according to the features in scalable bitstreams, we propose an optimal packetization scheme to minimize the impact of packet loss in packet erasure channels. We also design a novel and near-optimal packetization scheme to minimize the impact of bit errors in wireless channels. Secondly, we design an error-resilient unequal protection (ER-UEP) scheme to deal with packet loss efficiently and robustly. Compared with the conventional error protection schemes that suffer from bit contamination, our ER-UEP achieves much better error-resilient capability and can guarantee the successful decoding of all received bits. Thirdly, we introduce a new, efficient, and flexible method to detect errors in received bitstreams. Last but not least, a robust error concealment method is proposed for post-processing the reconstructed visual contents.