The template optimization of discrete time CNN for image compression and reconstruction

Abstract

Describes the A and B templates optimization of discrete time cellular neural network (CNN) for image compression and reconstruction. It is a very significant characteristic of CNN that nonlinear function and A template initiate some dynamics. Also, optimized B template contributes to the initial condition of some dynamics. This paper describes effectiveness by not only each A and B template but also the combination of A (dynamics) and B (filter) templates. It is a very significant point for CNN that the target issue is solved not by B template (filter) but by A template (dynamics) of CNN. The discrete time CNN with nonlinear quantization function proposed can encode (compress) images to small compressed code and can decode (reconstruct) its code to high quality lossy image. It is very important that the discrete time CNN state variable image which is determined dynamically based on the minimization of the discrete time CNN Lyapunov energy function to generate an optimized interpolative predict function is a lossy interpolative DPCM image between the original input and the interpolation predict functions.