In this paper

the aperiodic matrix model of deconvolution is described firstly

whose kernel matrix is a high order matrix. It's difficult to calculate this large matrix in practical image restoration problem directly by maximum entropy algorithm at a common computer for larger image. Second in this paper

we show a kind of deconvolution for overcoming the store problem of kernel matrix and use the conjugate gradient algorithm for the iterate convergence problem. The experimental results show its total computational burden and the memory requirement are moderate. It can run on a common PC and is suitable for many kinds of image blurring.