Research On Image Blur Detection And Blurred Region Segmentation

Image defocus blur or motion blur is due to the degradation caused by defocus of scenes or relative motion between scene and camera during the exposing time. Most of researches which aim at reducing blur effect to enhance image quantity regard image blur as a harmful phenomenon rather than take use of it. Blurred images usually contain depth information or motion information of the scenes, which can be used in multi-focus image fusion to extend depth of field or moving object detection and motion parameters estimation. The key to applying image blur information is to find a metric to evaluate image blur degree so that the blur region and clear region can be distinguished by the this metric. This paper studies on image blur detection, proposes and improves some image blurred region detection and segmentation algorithms. The main work is as follows:Firstly, by analyzing and studying the existing image blur detection algorithms, an image defocus blur region detection method based on correlation coefficient and local standard deviation is proposed. Pixels of defocus blurred image are classified into blur region or clear region by firstly computing the correlation coefficient and local standard deviation of their neighborhoods, which describe each pixel’s blur magnitude, and then comparing them to preset thresholds. Experimental results show that this method can detect blurred regions in the defocus blurred images efficiently.Secondly, by analyzing and studying the application of the singular value decomposition in image blur, an improved image blur region detection and segmentation method based on singular value decomposition and image matting techniques is proposed. Image patches that decomposed from a partially blurred image with suitable size are filtered by a same low-pass filtering, then a patch is marked as blurred or clear because of the singular value feature of clear patch having a bigger change than blurred patch after filtering. After that, a alpha matting technique is combined to exactly extract the blurred region. Simulated results show that the proposed method can accurately detect and extract blurred region from a partially blurred image.