Prediction of Adaptive Wavelet Packet Tree Structures for Signal and Image Coding
Corey Graves, North Carolina State University
Recently Wavelet Transform based compression techniques have demonstrated, superior performance over Fourier (and other traditional) Transform based compression techniques in terms of high quality, low bit-rate compression, for many types of signals. Even more recently it has be shown that Adaptive Wavelet Packets (variant of the Wavelet Transform) based techniques have the promise of greatly outperforming the standard Wavelet Transform techniques. A potential hindrance to the use of the Adaptive Wavelet Packet techniques, in very low bit-rate applications, is the amount of overhead information needed to encode what is the “best” wavelet packet tree structure for processing each signal segment (block) in a sequence of many signal segments (blocks) to be processed. For example, in the compression of 2-D image blocks, if 4-level Wavelet Packet Tree Structures are utilized, then, 66 bits of overhead information are needed. For 5-level tree structures, 261 overhead bits are needed. The number of overhead bits (along with the versatility of the compression algorithm) increases greatly with depth of the tree structures considered. We have developed an Adaptive Wavelet Packet method that avoids sending this overhead information all together. We call this the Predictive Adaptive Wavelet Packet method. Initial results from this research indicate that the new Predictive Adaptive Wavelet Packet (PAWP) method can achieve a rate/distortion performance that is better than that of regular Adaptive Wavelet Packet (AWP) methods, in which overhead information is needed. The trade-off is that about twice as much computation is required for PAWP as for AWP. Results have been acquired for speech signals and image sequences.
Abstract Author(s): Corey Graves