In video compression applications, a VLC (variable length code) is widely used to reduce statistical redundancy resident in given data. In most video coding systems such as MPEG-1,2, and 4, VLC tables are generally pre-defined and fixed for the sake of minimization of coding delay and VLC table retransmission overhead. However, the pre-determined symbol statistics does not properly represent the statistics of each video sequence, and local symbol statistics of a video sequence is also different from the global ones and it varies over time. Therefore, different VLC tables for each video sequence at a certain moment are required in order to effectively incorporate symbol statistics for improving coding efficiency. Though several algorithms concerning the code tree update were proposed, they are all mainly based on the text/data file compression and the estimation of symbol statistics. The better estimation, the higher efficiency we can obtain. However, due to the causality of video coding systems, we can only estimate the future statistics by the past statistics. Therefore, reasonable and efficient analysis of the past statistics is essential for better estimation.
In this paper, we propose an update technique using the past statistics of symbols within a finite frame window for estimating the statistics of symbols. The window buffer with frame memory size ‘n’ is used to store the most recently processed symbols and we update the VLC table based on them.
But in poor transmission environment, one drawback of proposed technique is the possibility of error propagation. An error in the bitstream may cause discrepancies between the tables at the receiver and the transmitter, thus resulting in synchronization failure. In this paper, we also propose an update technique minimizing this error propagation by using the multiple tables and the index of a table. Therefore we also propose the offline method that creates multiple tables and the online one that select one table among them.
The simulation results show that the technique using a window has better performance in coding efficiency compared with other algorithms and the technique using multiple tables gives a reliable method that create tables and select a suitable table.