Binary shape mask representation for zerotree-based visual object coding

摘要

A new technique for representing the binary shape mask of an arbitrarily shaped visual object is presented. By performing the shape-adaptive discrete wavelet transform (SA-DWT) on the object using a globally uniform subsampling policy, the shape mask is naturally decomposed via the "lazy wavelet transform". The resulting shape mask coefficients are arranged into spatial orientation trees for the purpose of parallel coding with the object texture in a zerotree-based scheme. This arrangement provides an excellent means for the prediction of uncoded shape pixels, thus lending itself to highly efficient coding.