Thesis (Ph. D.)--University of Washington, 2001. / Vita. Includes bibliographical references (leaves 242-249).

Video content is expected to account for 80 percent of all Internet traffic in 2019. There is therefore an increasing need for better video compression methods, to decrease the use of internet bandwidth. One way of achieving high video compression is to predict pixel values for a video frame based on prior and succeeding pictures in the video. The H.265 video compression standard supports this method, and in particular makes it possible to specify in which order pictures are coded, and which pictures are predicted from which. The coding order is specified for Groups Of Pictures (GOP), where a number of pictures are grouped together and predicted from each other in a specified order. This thesis evaluates how the GOPs should be structured, for instance in terms of sizing, to maximize the compression efficiency relative to the video quality. It also investigates the effect of multiple reference pictures, a functionality that enables the picture that renders the best prediction to be selected. The results show that the largest tested GOP size of 32 pictures is preferable for all tested video characteristics, and that support for multiple reference pictures renders a similar increase in compression efficiency for all GOP sizes.