Research On Local Blurred Area Detection In Image

During the acquisition process of digital images,if the imaging equipment focuses on the local scene or there is relative movement between the imaging equipment and the local scene,partial defocus blur or partial motion blur will appear in the imaging image.Detecting the blurred regions in an image accurately is the prerequisite for further extraction and utilization of image information.Therefore,blur detection research for partial blurred images has important academic significance and application value.At present,in most of blur detection algorithms,the sharp regions with flat texture in an image are easy to be misjudged as the blurred regions and the edge contours can’t be located accurately enough in detection results.In view of the above question,The detection of defocus blur and motion blur in an image is mainly focused in this thesis.(1)For the detection of defocus blur,a detection algorithm based on graph regularization optimization is proposed.The algorithm takes superpixel as the detection unit,fuses local binary pattern feature and region contrast feature to perform preliminary blur detection,then uses the graph regularization method to optimize the detection results.Compared with similar algorithms,the proposed algorithm has improved in terms of detection accuracy and speed.(2)For the estimation of defocus blur,the graph regularization method is used for the spread of blur information of the edge pixels,which effectively speeds up the blur estimation.(3)For partially motion blurred images with sharp foreground and blurred background,a detection algorithm based on singular value decomposition feature and image saliency constraint is proposed.The algorithm introduces image saliency into blur detection,combines effective information in blur map and saliency map by constructing a trimap,and uses the matting method to realize effective detection of partial blur.The algorithm can accurately determine the regions with flat texture and has high accuracy of edge localization.(4)For partially motion blurred images with sharp background and blurred foreground,a detection algorithm based on discrete cosine transform coefficient features is proposed by combining the gradient and boundary prior information about the image.The algorithm first uses discrete cosine transform coefficient features to detect regions with rich texture,and then combines edge gradient magnitude priors and boundary priors to distinguish regions with flat texture.Experimental results show that the algorithm can effectively detect motion blurred regions in images.