Mitigation of the effects of turbulence in the imaging path

Abstract

This thesis focuses on determination and correction of the information about disturbances obtained from images taken through a turbulent medium. The more distorted the images of a scene, the more corrections related to disturbances can be utilised to obtain unperturbed images. Two image restoration approaches explored in this thesis are: the Bispectrum algorithm and Lucky Region method. Also, an Image Alignment method is used as a pre-processing approach for correcting random shifts in an image caused by refractive-index variations in a turbulent atmosphere. The improved Bispectrum algorithm is applied in association with a speckle imaging method. Speckle imaging methods process a large number of short-time-exposure images. The turbulence in each of the short-time-exposure images is considered a frozen pattern. In the work of Wen [2], on which this work is based, there was an outstanding question of what sub- region size to choose to gain the preferable result. This work investigates this further in an empirical fashion, and examines whether the choice is image independent. A further examination of its properties shows that, when a 16×16 sub-region size is used with taking around 80 image frames, the best image restoration result is obtained. The Lucky Region method attempts to select the higher-quality regions from all image frames based on the value of Bicoherence, an image quality indicator, introduced by Wen [2] and shown to be effective for Lucky Region selection. An investigation of the range of ratios needed for the threshold in this method is provided. The best restored image occurs for threshold values defined between 10% and 20% of the frames with a 64×64 sub-region size. To achieve a better image reconstruction result, Image Alignment, which seeks to reduce the effect of the scene movement in an image sequence to prepare a better database for further image operations, is introduced for application in association with the Bispectrum and Lucky Region method. These combined image restoration algorithms demonstrate better performances than the direct approach of Wen’s original algorithms and are considerably more computationally efficient.