Stray Light Reduction Algorithms Based On Sparse Representation Theory

When a satellite is operating at midnight during the equinox, the sunlight may enter the field of view of the satellite camera. The radiation of sunlight reduces image quality. A portion of the field of view is close to blinding status. Strong stray light causes imaging time and coverage area degradation. The thesis presents two stray light suppression algorithms for a satellite camera based on Sparse Representation theory.Sparse Representation theory is widely used in image processing and machine learning. Sparse Regular Constraint, which is evolved from Sparse Representation, can also solve many image analysis and image processing problems.A stray light suppression algorithm based on Sparse Regular Constraint is presented in the thesis. It is known that stray light distributes smoothly in the space. Suppose the brightness surface of the degraded images which are interfered by the stray light has a strong correlation with the stray light distribution, the changes of the brightness surface of the degraded images can be used to approximate the real stray light distribution. An iterative algorithm is used to approximate the real stray light distribution by designing an energy function with two Sparse Regular Constraint to correct the degraded image. Experiments show that the algorithm can effectively estimate the distribution of stray light, resulting in good stray light suppression, and has strong adaptability.For large inputs, the stray light suppression algorithm based on Sparse Regular Constraint is not very efficient. Thus, we present a multiscale stray light suppression algorithm. The algorithm downsamples the degraded images into multiple scales, perform iterative computations on each scale and set the result of the previous scale as the initial value of the next scale. Split Bregman algorithm is used to improve the running time. Multiscale stray light suppression algorithm is much more efficient than the existing algorithms. Experimental results show that this algorithm can effectively improve the quality of degraded satellite images affected by stray light.