Noncausal image modeling with applications to image restoration

The problem of noncausal image modeling and estimation with applications to image restoration is considered. The noncausal vector autoregressive model for the image process is arranged into a descriptor system. This system is then decomposed into forward and backward stable subsystems which capture the dynamic behavior of the original descriptor system. The design of a time-varying filter is considered. The forward state is estimated, with the backward state considered as a colored output noise. The backward state is estimated, with the forward state regarded as colored output noise. Simulation results for noncausal image modeling are presented