Removing reflection from image using ICA
Matin Sadeghi, Masoud Reza Aghabozorgi Sahaf, Mohammad Taghi Sadeghi · 2008
When we view a scene through transparent glass or take a picture through a window or a windshield, usually the image consist of two linear superposition of images, a real image of the scene beyond the window observed through a glass and a virtual image of the scene reflected by the window on it. Decomposing the single input image from reflection is a massively ill-posed problem: in the absence of additional knowledge about the scene being viewed it may cause error in image processing for image analysis and measurement. In this paper we describe an algorithm that uses Basis Vectors obtained from ICA method to perform the decomposition, and then using coefficient of basis vectors and preserving the massive coefficients, perform the reconstructing real image. In the next step, to obtain original image we calculate correlation between estimated real image and images bank. Then we find exactly original image among the data base. By considering original real image as the second observation, we extract reflected image from the mixed image.