Identification of blur support size in blind image deconvolution
Li Chen, Kim–Hui Yap · 2004
This paper proposes a new approach for identifying the support size of the blur operator based on filter-response correlation criterion. As point spread function (PSF) is usually unknown a priori, blur identification becomes a fundamental issue in blind image deconvolution. The blur identification consists of two issues: (i) estimation of the support size, and (ii) computation of the coefficients. If the estimated blur support size differs from the actual support, the blur coefficients cannot be identified reliably. The proposed method addresses this problem through autocorrelation of the filtered image. The filter is derived from the degraded image using autoregressive (AR) model. By separating blur identification from image restoration, our method greatly improves the performance of the blind deconvolution process. Experimental results show that the method is effective in identifying blur size, further leading to satisfactory blind image deconvolution.