The paper presents an image deblurring method based on shrinkage functions in DCT domain. Themethod learns directly from example image patches. The functions can be approximated by a polynomialinterpolation. Then least square (LS) technique is used to optimal compute these shrinkage functions in discretecosine transform domain. The shrinkage functions can be applied directly to the other real images. The benefitsof the technique are eliminating the need for complex imaging modeling and statistical image priors. Themethod is the faster than other modeling ones.