Automated video enhancement from a stream of atmospherically-distorted images: the lucky-region fusion approach

An automated video enhancement technique capable of image fusion from a stream of randomly-distorted images of a still scene is presented in this paper. The technique is based on the "lucky-region" fusion (LRF) approach and aims to improve locally the image quality according to the following steps: (1) for each image of the video stream an image quality map (IQM) which characterizes locally the image quality is computed, (2) each IQM is compared to that of the current fused image leading to the selection of best quality regions (the "lucky-regions"), and (3) the selected regions are merged into the fused video stream. While the LRF approach succeeds in producing images with significantly improved image quality compared to the source images, its performance depends on the imaging conditions and requires adjustment of its fusion parameter - the fusion kernel size - in order to adapt to an evolving environment (e.g. a turbulent atmosphere). Parameter selection was so far performed manually using a trial-and-error approach which causes the technique to be impractical for a real world implementation. The automated LRF technique presented is relaxed from this requirement and selects automatically the fusion parameter based on the analysis of the source images making it more suitable for practical systems. The improved LRF technique is applied to imaging through atmospheric turbulence for various imaging conditions and scenes of interest. In each case automatically-fused video streams demonstrate increases in image quality comparable to that obtained with manual selection of the fusion parameter.