CNN image compression and reconstruction based on non-orthogonal wavelet transform

Abstract

In a pmctieal image processing such Wavelet Tmnsfonn (WT), the function orthogonality is required for reconstruction of the original image. The orthogonality has disadvantage that the selected filter is not necessarily optimal from a viewpoint from human retinal realization. It is not necessary to select an orthogonal templates in Cellular Neuml Network (GNN) image processing, because the CNN is non-linear analog circuit to obtain equilibrium points automatically and simultaneously. This paper describes CNN image compression and reconstruction based on a non-orthogonal WT. This system have an advantage of non-dependency of image scanning by spatio-tempoml GNN dynamics. It is very important that the rewnstruction of tmnsmitted compression image is done simultaneously by parallel neurons based on the "regularization" of ill-posed problem which is caused in a retinal system of a human bmin.