Enhancing the ability of NAS-RIF algorithm for blind image deconvolution

Structural adaptive anisotropic NAS-RIF for biomedical image restoration

“…where Dsup and Dsup are sets of pixels inside and outside the support region and LB is the background intensity. Over decades, generic NAS-RIF has undergone several improvements and extension [17][18][19][20][21][22][23]. A few years after it was introduced, NAS-RIF was enhanced by multiplying a DC gain of the FIR [17] to background value in the regularization, to ensure proper scaling during optimization.…”

“…Kundur et al extended NAS-RIF and proposed a fusion-based approaches, to simultaneously perform deblurring by using NAS-RIF and classifi cation by using Markov Random Field (MRF) [18]. To improve its robustness, a study [19] proposed modifying original cost function and incorporated an inter-band prediction. By using Haar wavelet decomposition, only a lower sub-band (LL 0 ) was restored with NAS-RIF, while the true image was then recovered from inter-band prediction.…”

“…The method has been widely applied in medical and astronomical imaging. Although several extensions have since been attempted to improve the convergence and quality of the restored image [17][18][19][20][21][22][23], enhancing of NAS-RIF with state-of-art algorithms remains an open area of investigation. This paper thus proposes a structural adaptive image restoration based on NAS-RIF algorithm.…”

Regularized Restoration Using Image Fusion for Digital Auto-Focusing

“…Performance evaluation in terms of PSNR and RMSE of the proposed algorithm with several existing restoration-based methods is shown in Table II. The RIR [2], RRM [3], RCLS [4], BER [7], discrete wavelet decomposition (DWD) [23], and pyramid decomposition (PD)-based methods [25] were implemented for comparison with the proposed method. The table also presents the simulation conditions for each method.…”

“…Consequently, the in-focused regions tend to have inaccurately restored artifacts, such as ringing or noise clustering. The nonnegativity and support constrained recursive inverse filtering (NAS-RIF) [7], for example, involves minimization of cost function, while nonlinear interpolative vector quantization (NLIVQ) and autoregressive moving average (ARMA)-based methods [8], [9] are borrowed from data compression technology. Although both the ARMA model and nonparametric approaches can provide acceptable focusing results under a set of assumptions, computational complexity due to numerical optimization and nontrivial number of iterations for convergence makes the corresponding system unrealizable as a commercial product.…”