Color image compression: Multiwavelets vs. scalar wavelets

Recently developed balanced multiwavelets have proven effective for image compression; however, their performance falls slightly short of the scalar wavelets. A previous comparison of some scalar wavelet and multi wavelet properties illustrated their importance for grayscale image compression. In this paper we present another property: the perfect reconstruction (PR) property; it too appears to explain the performance differences. This paper compares the compression performance of 3 scalar wavelets and 5 balanced multiwavelets on 7 color images. We observe that SA4 depicts the best performance amongst the multi wavelets in terms of both subjective quality and PSNR; this matches our expectation based on our results that indicate that it is the only BMW which satisfies the PR conditions. Although SA4 rivals the biorthogonal scalar wavelets in terms of many desirable properties, it suffers from a lower balancing/vanishing order. This difference explains the 0.01–0.47 dB performance gap between the best scalar wavelet and the best balanced multiwavelet.