Blind Restoration Methods Of Motion Blurred Image

For a imaging system, the relative movement between the camera and the objectoften lead to a motion blurred image, which seriously affected the quality andapplication of image. Generally, the Point Spread Function (PSF) of imaging system isunknown. Under this situation, we should estimate both the PSF and the real image, andwe name this problem as blind restoration. Blind restoration methods can be dividedinto two groups, one is priori blur identification method and the other is jointidentification method. Priori blur identification method should estimate the parametersof the PSF (for motion blur, they are blur direction and blur scale) firstly, and then usethis estimated PSF to restore the image. Joint identification method estimates the PSFand restores the image simultaneously. In reality, the information of the PSF is notcompletely unknown but partly known, in this case the restoration is a problem betweenclassical restoration and blind restoration, named as semi-blind image restoration.By analyzing the basic restoration theory of motion blurred image, this papermainly studied the blind restoration methods of motion blurred images, including a jointidentification method and a priori blur identification method. For semi-blind imagerestoration problem, a parameter estimation algorithm based on cross iteration forvariation model is put out. Specific works are as follows:1. For the method of joint identification, an improved Expectation Maximization(EM) restoration method for motion blur is put out by combining the non-local means(NLM) denoising algorithm and the EM algorithm, which can avoid the shortcoming ofsensitiveness to noise of EM algorithm. Numerical results showed that for the blurredimages with heavy noise, the improved EM could get higher Signal to Noise Ratio(SNR) than classical EM algorithm.2. For the method of priori blur identification, a new processing flow based on theprior of spectrum and cepstrum is put out. By adding the process of ringing suppressionafter restoration, the flow of the processing is more perfect, and the ringing caused byrestoration can be suppressed effectively. By setting a threshold to detect the line featurein spectrum or cepstrum of image, the estimation of blur direction is more accuracy andit will not be affected by the noise of background.3. For the problem of semi-blind restoration, a variation model is established, and across iteration algorithm based on split Bregman is put out to solve this model. By thisnew model, the parameter in PSF and the real image can be estimated simultaneously.The theory analysis shows that this method is also suitable for other blur kernels.