Method to detect and calculate motion blur kernel

Motion during camera's exposure time causes image blur, we call it motion blur. According to the linear system theory, if we can find the blur kernel which has the same meaning of point spread function, the blurred image can be restored by the blur kernel using iterative algorithms, such as R-L (Richardson-Lucy). Performance of the restoration is deeply depended on accuracy of the estimated blur kernel. In this paper we provide a novel method to detect and calculate the blur kernel. The process of kernel estimation can divide into two steps: The first step is detection of the motion path during the exposure time. A high-speed camera rigidly connected with the primary camera is used to capture a sequence of low resolution images, which contain information of camera position. While displacements of those images are detected, motion path can be drawn up. In the second step, blur kernel is calculated from the motion path by a novel model provided by this paper. Finally the blurred image captured by the primary camera can be restored by the kernel. We implement a hybrid imaging system for demonstration, and the experimental results prove the effectiveness of our method.