The dynamic behavior of the Internet's transmission resources makes it difficult to provide perceptually good quality streaming video. Scalable video encoding techniques have been proposed to deal with this problem. However, an encoded video generally exhibits significant data rate variability to provide consistent visual quality. We are, therefore, faced with the problem of accommodating the mismatch between the available bandwidth variability and the encoded video variability. We investigate quality adaptation algorithms for scalable encoded variable bit-rate video over the Internet. Our goal is to develop a quality adaptation scheme that maximizes perceptual video quality by minimizing quality variation, while at the same time increasing the usage of available bandwidth. We propose an optimal adaptation algorithm and a real-time adaptation algorithm based on whether the network conditions are known a priori. Experimental results show that the real-time adaptation as well as the optimal adaptation algorithm provide consistent video quality when used over both TCP-friendly rate control (TFRC) and transmission control protocol (TCP).

The heterogeneity of the Internet's transmission resources and end system capability makes it difficult to agree on acceptable traffic characteristics among the multiple receivers of a multicast video stream. Three basic approaches have been proposed to deal with this problem: 1) multicasting the replicated video streams at different rates; 2) multicasting the video encoded in cumulative layers; and 3) multicasting the video encoded in noncumulative layers. Even though there is a common belief that the layering approach is better than the replicated stream approach, there have been no studies that compare these schemes. This paper is devoted to such a systematic comparison. Our starting point is an observation (substantiated by results in the literature) that a bandwidth overhead is incurred by encoding a video stream in layers. We argue that a fair comparison of these schemes needs to take into account this overhead, as well as the specifics of the encoding used in each scheme, protocol complexity, and the topological placement of the video source and the receivers relative to each other. Our results show that the believed superiority of layered multicast transmission relative to replicated stream multicasting is not as clear cut as is widely believed and that there are indeed scenarios where replicated stream multicasting is the preferred approach.

Dynamic behavior of the Internet's transmission resources makes it difficult to provide perceptually good quality of streaming video. MPEG-4 Fine-Grained Scalable coding is proposed to deal with this problem by distributing the data in enhancement layers over a wide range of bit rates. However, encoded video also exhibits significant data rate variability to provide a consistent quality video. We are, therefore, faced with the problem of trying to accommodate the mismatch between the available bandwidth variability and the encoded video variability. In this paper, we investigate quality adaptation of the layered VBR video generated by MPEG-4 FGS. Our goal is to develop a quality adaptation scheme that maximizes perceptual video quality through minimizing quality variation while at the same time increasing the usage of available bandwidth. We develop an optimal adaptation scheme and an online heuristic based on whether the network conditions are known a priori. Experimental results show that the online heuristic as well as the optimal adaptation algorithm provide consistent video quality when used over both TFRC and TCP.

We introduce an overlay network architecture and signaling mechanism that permit program insertions in live, high quality video streams transmitted over IP networks. We describe the implementation of an application proxy that dynamically inserts pre-recorded video programs into NTSC D1 quality Motion-JPEG streams with no visible artifacts. As increases in computing power further enable the modification of video during transport, new services such as personalized commercial advertisement insertions promise to make IP distribution an attractive alternative to conventional broadcast network distribution.

The heterogeneity of the Internet's transmission resources and end system capability makes it difficult to agree on acceptable traffic characteristics among the multiple receivers of a multicast video stream. Three basic approaches have been proposed to deal with this problem: 1) multicasting of replicated video streams at different rates, 2) multicasting the video encoded in cumulative layers, and 3) multicasting the video encoded in non-cumulative layers. Even though there is a common belief that the layering approach is better than the replicated stream approach, there has been no studies that compare these schemes. This paper is devoted to such a systematic comparison. Our starting point is an observation (substantiated by results in the literature) that a bandwidth penalty is incurred by encoding a video stream in layers. We argue that a fair comparison of these schemes needs to take into account this penalty as well as the specifics of the encoding used in each scheme, protocol complexity, and the topological placement of the video source and the receivers relative to each other. Our results show that the believed superiority of layered multicast transmission relative to stream replication is not as clear cut as is widely believed and that there are indeed scenarios where replication is the preferred approach.