| Summary: | This paper addresses the issue of non-uniform motion deblurring for a single photograph. The main difficulty of spatially variant motion deblurring is that, the deconvolution algorithm can not directly be used to estimate blur kernel, due to the kernel of different pixels are different with each other. In this paper we firstly build up the camera pose space, and take the blurred image as the weighted summation of all possible poses of the latent image. Then the deblurring problem is converted to searching for the optimized weighted parameters in the pose space. Due to its high dimension and non-convexity we propose a framework using the particle swarm optimization algorithm to solve the problem iteratively. We also find that regions with high frequency texture may damage the deblurring process, which motivates a new latent image prediction method. A non-linear structure tensor with anisotropic diffusion and a shock filter are combined to smooth the image while keeping the salient edges of it. Experimental results show that our approach makes it possible to model and remove non-uniform motion blur without hardware support.
|
|---|