Research On Motion De-blurring From A Single Image

With the popularization of digital photography,image enhancement has become a hot spot in the field of digital image processing.As one of the branches of image enhancement,de-blurring technique has been paid more and more attention.According to the blur kernel which causes image blurred,the de-blurring technique is divided into defocusing de-blurring and motion de-blurring.Image blurred occurs due to camera shake or rapid movement of the subject within the exposure time of the camera if camera focus fails or the subject's thickness exceeds the depth of field in the imaging scene.This thesis focuses on motion blurred images.The mathematical model of the motion blurred image is generally a convolution of the clear image with the blur kernel plus the random noise.Therefore,to the motion de-blurring technique is to estimate the blur kernel from the blurred image,and de-convolute to get a clear image.According to the different properties of blur kernel,it can be divided into two categories:space invariant and space variant,and their mathematical models correspond to single core model and multi-core model respectively.Space invariant motion de-blurring technique assumes that the blur kernel in the whole image is consistent,and the space invariant motion de-blurring technique is not monotonically invariant with the different pixels.It can be seen that the mathematic model of the space invariant blurred image is more complex,and its de-blurring technique will be more challenging.In this thesis,all kinds of motion de-blurring methods are researched and analyzed.Combining with the research hotspots and difficulties in the field of de-blurring,the motion blurred images with variant space is this thesis's research focus on.Therefore,this thesis presents a new framework for motion-blurred images with variant space-based motion detection based on blur detection.On the basis of the new framework,the first is to detect the blurred region in the local blurred image,and then the blurred region is segmented according to the detected blurred region map.Finally,the clear image of the blurred region is obtained by using the space invariant motion de-blurring technique for the segmented blurred regions.The main work of this thesis is the four key steps:(1)The blurred region of the local blurred image is detected by using the features of the blurred image;(2)The blurred region is segmented by using the corresponding blurred map;(3)Due to the blur kernel is uniform in each region of the image,using the advanced spatial invariant de-blurring technique to get the clear region;(4)The use of edge expansion to eliminate the regional splicing edge effect.In both the artificial data set and the real blurred image dataset experimental results verify the effectiveness of the proposed method.In the artificial data set,the PSNR is improved by lOdB and the RMSE is reduced by 0.12 compared with the traditional single-core antialiasing algorithm.In the real blurred image data set,compared to other algorithms visual effects have significantly improved,the ringing effect is also significantly inhibited.Finally,the experimental results show that the new method is more effective than the single-core-based motion blur algorithm.Compared with the multi-core algorithm,the algorithm model is obviously simplified,and the computational complexity is also significantly reduced.