A new approach for estimation of wavelets with non-separable kernel from a given image

In this paper, we propose a new approach for the estimation of wavelet with non-separable kernel from a given image. The paper discusses method to the design of biorthogonal 2D-FIR perfect reconstruction filter banks where the filters are estimated from a given image. The design method is proposed with different sampling lattices. The important contribution of this paper lies in the fact that these filters are estimated from given image itself. Several design examples are presented to show the efficacy of the approach and also the ease of design. Since these wavelets are estimated from given image, thus they are called 2-D matched wavelets.