A Multi-Constrained Iterative Algorithm For Image Restoration

Image degradation refers to the image distortion during the processes of acquisition, transmission, processing, and so on. It is caused by external environments or hardware devices. We need to restore the degradation image effectively before it can be used in other fields. Image restoration technology is widely used in various fields of society, and it is very important for our life.Blur and noise are the two main forms of image degradation. The degrade image with image blur and noise is also a hotspot in the research of image restoration. This thesis is focus on the two degradation models. One is the degradation that with motion blur and Gaussian noise and another is the defocusing blur-impulse noise model.For the image degradation that is corrupted by motion blur and Gaussian noise, this thesis uses the TV-POCS-KSVD model to recover the image. On the basis of traditional TV-POCS model, we add the image sparse representation constraints. This thesis designs a contrast experimental to verify the validity of this method. The experimental results show that TV-POCS-KSVD algorithm can remove the Gaussian noise at the same time retain the edge of the image. And experimental results show that the new method has the better restored effect than the TV-POCS model. To solve the above optimization problem, this thesis uses the split Bregman method in which the problem is transformed into multiple sub-optimization problems.For the detection of impulse noise, this thesis introduces an improved boundary detection algorithm. The experimental results show that the proposed algorithm can effectively detect the “salt-and-pepper” impulse noise and Random-valued impulse noise. And the accuracy of the detection is higher than the AMF, BDND and the other traditional methods. In this thesis, we adopt the two-step restoration model to recovery the degradation image that is corrupted by defocusing blur and impulse noise. The first phase of this method is to identify the possible impulse noise positions using the improved boundary detection algorithm; the second phase is to recover the image via the TV-POCS-KSVD model. This thesis designs a experimental to verify the validity of this method. Experimental results show good result.