Super-Resolution of Unmanned Airborne Vehicle Images with Maximum Fidelity Stochastic Restoration

Super-resolution (SR) refers to reconstructing a single high resolution (HR) image from a set of subsampled, blurred and noisy low resolution (LR) images. One may, then, envision a scenario where a set of LR images is acquired with sensors on a moving platform like unmanned airborne vehicles (UAV). Due to the wind, the UAV may encounter altitude change or rotational effects which can distort the acquired as well as the processed images. Also, the visual quality of the SR image is affected by image acquisition degradations, the available number of the LR images and their relative positions. This dissertation seeks to develop a novel fast stochastic algorithm to reconstruct a single SR image from UAV-captured images in two steps. First, the UAV LR images are aligned using a new hybrid registration algorithm within subpixel accuracy. In the second step, the proposed approach develops a new fast stochastic minimum square constrained Wiener restoration filter for SR reconstruction and restoration using a fully detailed continuous-discrete-continuous (CDC) model. A new parameter that accounts for LR images registration and fusion errors is added to the SR CDC model in addition to a multi-response restoration and reconstruction. Finally, to assess the visual quality of the resultant images, two figures of merit are introduced: information rate and maximum realizable fidelity. Experimental results show that quantitative assessment using the proposed figures coincided with the visual qualitative assessment. We evaluated our filter against other SR techniques and its results were found to be competitive in terms of speed and visual quality