Compression with Time-Frequency Localization Filters

The wavelet transformations combined with the zerotree algorithm have proven to be an efficient compression scheme in many applications. This scheme keeps filters fixed at the all resolution levels. We propose here a bi-filter method that uses different filters at different resolution levels. At high levels, good frequency localization filters are used. They are replaced by good time localization ones at lower levels. We choose to use filters derived from interpolatory refinable functions. The detailed design and analysis are included. The numerical empirical results show that this bi-filter compression scheme keeps more texture details of the original images. §