Compression of quincunx subbands

Abstract

Quincunx downsampling splits the spectrum of an image in two diamond-shaped subbands. This allows the information to which the human eye is most sensitive to be preserved, and also requiring less bits in a lossy compression. We develop an algorithm that performs successive quincunx subband decompositions, thus allowing multi-level decompositions and efficient coding. Theory of lattices is briefly reviewed, and simulations using a simple subband coder were carried out. Results obtained are compared to the rectangular separable subband decomposition, showing that, without exploring redundancies due to quincunx symmetries, both strategies deliver similar results.