版权所有:内蒙古大学图书馆 技术提供:维普资讯• 智图
内蒙古自治区呼和浩特市赛罕区大学西街235号 邮编: 010021
作者机构:Shanghai Univ Sch Commun & Informat Engn Shanghai 200072 Peoples R China
出 版 物:《IEE PROCEEDINGS-VISION IMAGE AND SIGNAL PROCESSING》 (IEE Proc Vision Image Signal Proc)
年 卷 期:2006年第153卷第6期
页 面:825-836页
核心收录:
主 题:ELECTRONIC data processing IMAGE reconstruction IMAGE processing COMPUTATIONAL complexity DIGITAL images DECOMPOSITION method (Mathematics)
摘 要:Image restoration is formulated using a truncated singular-value-decomposition (SVD) filter bank. A pair of known data patterns is used for identifying a small convolution operator. This is achieved by matrix pseudo-inversion based on SVD. Unlike conventional approaches, however, here SVD is performed upon a data-pattern matrix that is much smaller than the image size, leading to an enormous saving in computation. Regularisation is realised by first decomposing the operator into a bank of sub-filters, and then discarding some high-order ones to avoid noise amplification. By estimating the noise spectrum, sub-filters that produce noise energy more than that of useful information are abandoned. Therefore high-order components in the spectrum responsible for noise amplification are rejected. With the obtained small kernel, image restoration is implemented by convolution in the space domain. Numerical results are given to show the effectiveness of the proposed technique.