|M.Sc Student||Diamant Yaron|
|Subject||Overcoming Secondary Reflections|
|Department||Department of Electrical Engineering||Supervisor||Professor Yoav Schechner|
|Full Thesis text|
An image acquired through a glass window is a superposition of two images: the scene behind the window, and a reflection of the scene in front of the window. When light rays hit the surface of a glass window, they are reflected back and forth inside the glass , causing many internal reflections, both for the scene behind the window and for the scene in front. Our work deals with the problem of those internal reflections. First, we present a physical model of the problem. It turns out that each scene undergoes a convolution with a point spread function (PSF), composed of distinct delta functions. Therefore, the solution for this problem involves inversion of this PSF. We analyze the fundamental limitations faced by any attempt to solve this problem. These include unknown boundary conditions and attenuated frequency components. Additional problems are estimation of the PSF and recovery when the distance between the delta functions of the PSF is not an integer number of pixels. Beside this analysis, we present an approach to solve this inverse problem (within the mentioned fundamental limitations). The approach is based on deconvolution by linear optimization, while the PSF parameters are automatically estimated. The inputs to the deconvolution algorithm are two frames. The frames are taken through a polarizing filter , which causes changes in the PSF. The result of the deconvolution
is separation of the scene behind the window and the scene in front, and elimination of most of the effects associated by the mentioned secondary reflections. However, residual artifacts remain. These artifacts are treated by nonlinear post-processing .
The method is demonstrated experimentally .