Description
This thesis proposes the fusion of multiple images, that have undergone image degradation due to atmospheric turbulence, in order to obtain a higher quality, single frame representation of the “true” image. While the correction of turbulence is nothing new, our focus is on utilizing the divergence of velocity flows to select the high quality regions to later be fused—a method with little current literature/research. This fusion process consists of two steps. The first is the non-rigid registration of the input image sequence to reduce any geometric distortions due to atmospheric turbulence. This is achieved through the use of optical flow methods, which give velocity flows at each registered location. The second step is the selection of high quality regions when fusing the images to obtain a sharper output image. We assume the sharp regions can be located by utilizing the divergence of the velocity flows that were recovered in step one. Throughout the explanation of this method we also assess optical flow and its importance not only in image registration, but also in the selection of fused regions. Primal-dual methods are also explained, since those are vital to solve the image fusion model.