Image compression based on restcted wavelet synopses with maximum error bound

Abstract

The construction of wavelet synopses with maximum error bound have been studied by many years, which has many real world applications, such as health data stream analysis and image compression. Recently, there are two kinds of wavelet synopses with maximum error bound: one is restricted wavelet synopses whose synopses are restricted to the haar coefficients. Another is unrestricted wavelet synopses whose synopses are not equal to the haar coefficients. In this paper, we propose a simple and novel compression approach of restricted wavelet synopses with maximum error bound. The approach is applied in image compression in this paper. This approach firstly performs a stepwise Haar transformation for each 2 X 2 non-overlapping sub-block of an image. It then filters each level of detail coefficients by using varied thresholds. This approach guarantees each pixel's error in an user-defined error bound and can maintain image quality greatly in reconstruction. Experiment results show that the approach has faster execution speed and can obtain better reconstruction effects under the same compression ratio than the existing approaches. This advantage is more obvious, especially in the situation of high compression ratio.