Adaptive Error Based Thresholding for On-Line Wavelet Compression

Abstract

Wavelet based data compression is superior to interpolative compression techniques like boxcar as wavelets provide an efficient joint time-scale representation of the signal in terms of wavelet coefficients. Thresholding over these coefficients is used to achieve compression. A recursive on-line wavelet based compression algorithm, which combines the high quality of compression of wavelets and on-line implementability of interpolative methods has already been developed. In this paper, an error based thresholding framework is introduced, which adaptively computes these thresholds, given the bounds on root mean square error and local point error. This framework is applied to the recursive on-line wavelet based compression using Haar wavelets. Experiments show that the resulting algorithm gives superior compression as compared to interpolative methods. Moreover, it can be used on-line and provides an effective way of controlling the local point error (LPE) and the root mean square error (RMSE).