Abstract
Distribution of video can be done by different means. A signal may be multicasted, where several eceivers can watch the same video stream, or distributed on a one-to-one basis. A distribution where one transmission may cover several receivers reduces the total bandwidth usage, while the strength of one-to-one distribution is that a receiver may control the flow of signals.
Receiving devices vary in screen sizes, processing capabilities, and available bandwidth. As cell phones have gained access to the Internet there has been many new techniques for displaying a web page well on both cell phones and computers. An equal ability to scale video streams is also a much wanted feature.
The Distributed Media Journaling (DMJ) project attempts to implement a prototype that generates one signal that is displayable for all receivers regardless of bandwidth, processing capabilities and screen sizes. The prototype makes use of a content-based network to transmit video signals.
For compression the prototype makes use of forward prediction and discrete cosine transform (DCT). The video signals are compressed considerable, but it is still not a very efficient compression mechanism. This thesis describes the addition of two compression techniques that will enhance the compression ratio of the prototype without lowering the quality of the signal. The techniques are called motion compensation and bidirectional coding.
The motion compensation algorithm shifts elements of a former frame in order to provide a frame that better matches a current frame. Most frames in the video are represented as a differential to another frame. If a better match can be found, less correctional data is needed, hence resulting in better compression.
Bidirectional coding contribute to compression by also allowing future frames to be used as a base for differential coding. In many cases better matches can be found in future then in former frames. Experiments made as part of this thesis shows that by applying motion compensation and bidirectional coding the size of the differential code in the video stream is reduced by up to 70%.