Deblurring of UAV aerial images for civil structures inspections using Mumford-Shah / Total variation regularisation

This paper addresses the problem of blind image deconvolution, in the context of a real-life application. The purpose of our study is to present a newly developed technique, and to test its deblurring capabilities on images acquired by an UAV (unmanned aerial vehicle) used for the civil inspection of bridges. We propose an approach based on the double regularisation principle, as introduced by You and Kaveh. Our scheme lies on the use of a total variation (TV) regularisation term on the blur kernel, and a Mumford-Shah functional based one, on the image. We give the details of the minimisation algorithm and provide some preliminary results both on synthetical and real images.