Automatic Reflection Removal Using Gradient Intensity And Motion Cues

When taking photos,we usually face the situation that there exists an obstacle between scene and photographers.Thus the image of photographer will be reflected to the photo.We present a method to separate the background image and reflection from two photos that are taken in front of a transparent glass under slightly different viewpoints.Methods before need a lot of images and priors,such as thickness of glass,distance of reflection,but our method only need two photos,which make the method can be used widely.In this paper,we introduce this method in several aspects,the motion score,the intensity score and reconstruction of image.In our method,the SIFT-flow between two images is first calculated and a motion hierarchy is constructed from the SIFT-flow at different levels of spatial smoothness.To distinguish background edges and reflection edges,we calculate a motion score for each edge pixel by its variance along the motion pyramid.Alternatively,we can group edge pixels in a same superpixel into edge segments and calculate the motion scores by averaging over each segment.In the meantime,we also calculate an intensity score for each edge pixel by its gradient magnitude.We combine both motion and intensity scores to get a combination score.A binary labelling(for separation)can be obtained by thresholding the combination scores.We construct optimization function whose condition is the binary labelling and the objection is the similarity with original image.The background image is finally reconstructed from the separated gradients.Compared to the existing approaches that require a sequence of images or a shot video clip for the separation,we only need two images,which can be taken from the stereoscopic camera with once exposure or the normal camera with twice exposure,which largely improves its feasibility.We use several class of images,such as the real image taken in the real world,the real image taken in the artificial scene and the synthesis image constructed by computer to test our method.Various challenging examples are tested to validate the effectiveness of our method.