Three-dimensional transform coding of partitioned video sequences with matching pursuits

This work introduces an alternative video coding approach that exploits the temporal correlation present in video signals by applying a combined wavelet/DCT three-dimensional transform to the input video sequence. As the correlation along the temporal axis is likely lowered when motion is present we first partitioning the input sequence into partitions or groups of similar pictures and then applying a transform to highly correlated groups of pictures. Although we apply the transform to correlated pictures, a fraction of the signal energy that is due to motion is scattered in a small number of non-zero "high-frequency" coefficients and is not captured by the compacted "low-frequency" coefficients. The remaining coefficients that are due to motion are then coded using the matching pursuit technique. The majority of the non-zero coefficients are coded by a well-known image coding technique — the LZC coder.