Quantification of distortion in long-range video sequences: Straightness metric and reference frame generation
Abstract
In software-based turbulence mitigation, reliable no-reference image quality metrics are indispensable. In this research, a straight line detector (SLD)1 is used to quantify warping due to turbulence. Our metric is based on the average length of detected straight lines, which constitutes a credible representation of the straightness in the image. The performance of the metric was tested on an anisoplanatic warping simulation. With increasing warping strength, the metric shows a decreasing average length of detected straight lines in the images. It is shown that the metric is able to differentiate between a set of simulated data with strong warping (C2n = 10-14 m-2/3) compared to weak warping (C2n = 10-15 m-2/3). Hereby the least and most warped frames found from the metric are considered and checked for credibility. A reference frame generation is developed by averaging the four least warped frames found from the metric. For evaluation of the reliability of the metric, it is tested on several data sets acquired at different distances. A correction of dewarping with optical ow is performed and the results with the reference frame generated based on the straightness metric and with a long exposure image as reference are compared.
