Error-Resilient Video Coding over IP Networks

Licentiate thesis, 2005

By its very nature, compressed data is more sensitive to bit corruption and data loss than uncompressed data. Transmitting data across unreliable bandlimited channels can therefore in many cases lead to an unsatisfactory reception of the source. This thesis deals with two distinct problems related to the transmission of video sequences across the Internet. The first problem addressed by this thesis, is the design of an error-resilient video coder for use on the Internet. A video coder is proposed which is based on the use of harmonic frames and leaky prediction coding. Harmonic frames are used to insert spatial redundancy into the compressed bitstream while leaky prediction coding is used to introduce temporal redundancy. The second focus of this thesis is channel prediction for the Internet. Congestion is the main cause of packet losses in the Internet. Congestion has previously been successfully predicted from measurements of the end-to-end variations in packet arrival times. Motivated by this, an algorithm is proposed to predict the probability of packet losses from the fluctuations in packet delays. The work was funded by Vinnova as part of the IPVideo project, a collaborative research project among Chalmers University of Technology, Linköping University and TopOneTech AB.