Restoration of atmospheric turbulent video containing real motion using rank filtering and elastic image registration

Atmospheric turbulence causes a random blur in images due to random fluctuations of the refraction index of the air through which the light propagates. An algorithm is described that compensates image distortion due to the atmospheric turbulence in video while keeping the real moving objects in the video unharmed. The result of the restoration of the video sequence is a much more 'stable' sequence where real moving objects are easier to detect and track. This algorithm is based on Differential Elastic Image Registration which maps each turbulent image to a one good reference image derived from the sequence of the turbulent images. To preserve real motion in the scene, the moving objects are located and the compensation for the distortion of the turbulence is applied only to the static areas of images. The algorithm provides also a smooth transition between the static and the moving areas in the image.