Complexity evaluation of random access to coded Multi-View Video Data

Significant progress has recently been made in the coding of video data captured with several cameras. Applying image-based rendering algorithms to this data, the user can watch the scene from any desired viewpoint. For that, the renderer may need arbitrary images from different camera views. That is why random access to the coded data is crucial. However, in efficient multi-view coding schemes, random access is complicated by the complex relations between the different frames. In this paper, the Joint Multi-View Video Model (JMVM) is exemplarily used to analyse the complexity of random access to multi-view video sequences. It is shown how the effort to access a certain frame depends on its position in the sequence, and the average and maximum complexity to access a randomly selected frame is quantified. The influence of the number of camera views and the length of each group of pictures is illustrated and discussed.