Window Image Reflection Removal by Blind Image Separation

概要

論文の詳細を見る
A novel second order statistics approach to blind image separation has been deployed to remove reflection image from window glass images. The reflection image and the image from behind the window glass are considered as blind sources. Since sources are uncorrelated, their low frequency as well as high frequency halfbands are uncorrelated too. It is shown that the un-mixing matrix is obtained by simultaneous diagonalization of covariance matrices of low and high halfbands of the mixtures. The proposed technique has been compared with AMUSE [1] and SOBI [2] as two other well known second-order statistics BSS techniques. Three different numerical metrics approves the efficiency dominance of the proposed technique.