Research On Blind Restoration For Interested Information In Blurred Image

In the areas of criminal forensics, judicial appraisal, intelligence gathering and transportation monitoring, the key information in the image is crucial. However, due to the existence of a variety of factors, the image would be blurred in its acquisition process, which makes the extraction of information from image difficult, and even affects the related work to carry on. Because of the great challenge of the image restoration technique and the complexity of degraded reasons of actual blurred images, most of the existing blind image deconvolution methods have only some theoretical value, which means they have not provided effective ways to solve practical problems and the practicability needs to be improved. The research on blind restoration for actual blurred images has a wide application prospect, and is one of research focus and difficulty in the field of image processing.This thesis makes an in-depth research on the blind restoration of interested information from single blurred image, by applying the combined methods of theory analysis, algorithm design and experiment validation. The main contents and innovations are as follows.Firstly, a blind restoration approach for the out-of-focus blurred image is studied. The out-of-focus blur is caused by the inaccurate focus of the imaging equipment. An image restoration algorithm is proposed to restore the specific type of blurred image. In this algorithm, the Canny edge detector and Hough transform are used to detect straight edges in the image. Then the step edges or approximate step edges are located, adopting a proposed method based on the step edge spreading statistical characteristic and the modified Grubbs method. The optimal area is extracted adaptively to calculate the line spread function based on the detected step-edges. The radius of the out-of-focus blur is then obtained adopting the relationship between the radius and the line spread function. Finally, as the radius is known, the point spread function is obtained, and the blurred image is restored by the non-blind deconvolution method. The experiment results show that the proposed algorithm can estimate the blur parameter accurately, and can be applicable to out-of-focus blurred images of single and multiple depth of field. Tested on the actual blurred images, the results show the algorithm can restore the interested information from blurred image better, and has higher practical value.Secondly, a blind restoration approach for the camera-shake blurred image is studied. Based on variational Bayesian, an improved algorithm is proposed for the camera-shake blurred image, which results from the camera shaking during the exposure time. In this algorithm, Using the prior statistical characteristic of the natural images gradient and point spread function, the point spread function is estimated through the variational Bayesian method, and the blurred image is restored by the non-blind deconvolution method. In order to improve the algorithm's efficiency and ensure the estimation accuracy, based on run-length statistics, a method is proposed to select the region of rich edges automatically, which is used to estimate the point spread function instead of the entire image. In addition, based on sub-region detection and weighed Fuzzy mean filter, a dering method is proposed to reduce ringing effect. Test on the actual blurred image, the experimental results show that the proposed algorithm can remove the camera-shake blur to a great extent, and can improve the algorithm's efficiency, while ensure the estimation accuracy of point spread function, and can reduce the ringing effect, while preserve the image edge and details well.Thirdly, a blind restoration approach for the partial-motion blurred image is studied. Based on the alpha channel of image, an image restoration algorithm is proposed for the partial-motion blurred image, which is caused by object's motion before static background. In the algorithm, the alpha channel of blurred image is calculated using image matting technique. The blur length and angle are estimated by a proposed multi-scale method based on alpha-motion constrain. Then based on the relationship between point spread function and alpha channel, a method is presented to obtain the point spread function, which is used to restore blurred image by non-blind restoration method. The final result is composed of recovered foreground and clear background by a designed method. The experimental results show that the proposed algorithm can estimate the blur length and angle accurately, and can estimate the point spread function exactly in the both conditions of uniform and non-uniform linearity motions. Accordingly, the interested information can be restored from blurred image effectively. At the same time, the proposed algorithm can compose the recovered foreground and clear background better, which improves the quality of the restored imageFinally, a blind restoration approach for overlap blurred image is studied. Based on a coded-shutter model, an image restoration algorithm is proposed for overlap blurred image, in which there exists overlap phenomenon. Motivated by the characteristic of overlap blurred image and the imaging principle of camera, a coded-shutter model is presented. As a first attempt, the degraded principle of the overlap blurred image is simulated by this model, and an image restoration algorithm is designed based on the model. The experimental results show that the coded-shutter model can simulate the nature of overlap blur effectively, and has a certain rationality for the restoration. Tested on the actual blurred image, it shows the proposed algorithm can restore the interested information from blurred image better, and has higher practicability.