Kernel Fusion for Better Image Deblurring
Long Mai and Feng Liu
Department of Computer Science, Portland State University

  The source image in this figure was used under a Commons Creative license from Flickr user vanherdehaage.
Abstract
Kernel estimation for image deblurring is a challenging task and a large number of algorithms have been developed. Our hypothesis is that while individual kernels estimated using different methods alone are sometimes inadequate, they often complement each other. This paper addresses the problem of fusing multiple kernels estimated using different methods into a more accurate one that can better support image deblurring than each individual kernel. In this paper, we develop a data-driven approach to kernel fusion that learns how each kernel contributes to the final kernel and how they interact with each other. We discuss various kernel fusion models and find that kernel fusion using Gaussian Conditional Random Fields performs best. This Gaussian Conditional Random Fields-based kernel fusion method not only models how individual kernels are fused at each kernel element but also the interaction of kernel fusion among multiple kernel elements. Our experiments show that our method can significantly improve image deblurring by combining kernels from multiple methods into a better one.
Paper
Long Mai and  Feng Liu. Kernel Fusion for Better Image Deblurring.
IEEE CVPR 2015, Boston, MA, June 2015. Full Paper Extended Abstract
Dataset
1. Original images used to create our dataset. These images are used under a Commons Creative license. This dataset  also contains an excel file and note.txt file that can be used to retrieve the urls of these images. (zip)

2. The 2976 blurry images along with their corresponding ground-truth images and the blur kernels used to create them. All images
are in gray-scale.. (zip)