|M.Sc Student||Averbuch Eugene|
|Subject||Adaptive Filtering of Visibility Degraded Images|
|Department||Department of Electrical Engineering||Supervisor||Professor Yoav Schechner|
|Full Thesis text|
When imaging in scattering media such as fog, haze and water, the visibility degrades as objects become more distant. A growing interest in the analysis of such images has led to the development of several methods that claim to restore good visibility. Indeed, visibility can be significantly restored by computer vision methods that account for physical processes occurring during image formation. Nevertheless, such recovery is prone to noise amplification in pixels corresponding to distant objects, where the medium transmittance is low. Prior recovery studies have hardly dealt with the noise problem. In this work we analyze the nature of this noise amplification. We then present an adaptive filtering approach that counters the above problems and is applicable to a variety of scenarios of visibility degradation due to scattering effect. The method we describe here greatly improves visibility relative to raw images, while inhibiting the noise amplification. Essentially, the recovery formulation is regularized, where the regularization adapts to the spatially varying medium transmittance. The proposed approach emphasize the regularization in pixels corresponding to low transmittance, and turns off the regularization in pixels corresponding to high transmittance. Thus, this regularization does not blur close-by objects. A small modification of the proposed approach allows using the distance-adaptive smoothing in image-based rendering of focus and defocus effects. We demonstrate the approach in experiments where the scene radiance and distance map are recovered in haze and underwater.