Structural adaptive anisotropic NAS-RIF for biomedical image restoration
Paramate Horkaew, Tanawat Kwanpak · Istrazivanja i projektovanja za privredu · 2019
Blind image deconvolution is an ill-posed problem that attempts to restore an acquired image degraded by unknown PSF. A variational BID implementation, called NAS-RIF, is known for being robust but prone to poor convergence under low SNR and unrealistic support. Motivated by simple yet efficient fidelity metric, this paper presents an improved NAS-RIF by reducing adverse effect of inverse high-pass filter and computationally intensive pre-deterministic noise removal, by adaptively incorporating anisotropic structural property within local neighborhood seamlessly in NAS-RIF cost function. With an automatic support region estimation, the entire deconvolution process was fully automatic. The experimental results reported herein indicated that the enhanced structural adaptive anisotropic NAS-RIF had better convergence condition, while maintaining the underlying image fidelity.