Blur Detection Based On A Theoretically Guaranteed Indicator Function

Motion blur is very common in our daily life. If there is a movement like fast moving car or running man in the scene when taking a picture, there will be motion blur. In the area of image restoration, if we want to remove the motion blur in an image, we must segment the blurred area first. Motion blur detection must to be done before motion blur segment. In addition, motion blur detection can provide some useful information for moving object detection in a single image. Therefore, motion blur detection in a motion blur image is very necessary. A new motion blur detection algorithm different from others is put forward in this thesis after a further study of motion blur. Not only this algorithm can get the same effect with current best algorithm, but also can reduce the computing time greatly. The complexity of the new algorithm is very low, and can achieve real-time almost.Motion blur detection based on mixture Gaussian model algorithm is introduced firstly in this thesis, and the reason for inefficiency is pointed out. In order to solve this problem, a new blur indicator function is proposed in this thesis. The theorem of blur separability of this indicator function is given in particular. This theorem guarantees the separability of motion blur, and lay a theoretical basis for blur detection. At last, the experiments on three public databases verify the effectiveness and advantages of the proposed algorithm.