Recovering Badly Exposed Objects From Digital Photos Using Internet Images

In this paper we consider the problem of clipped-pixel recovery over an entire badly exposed image region, using two correctly exposed images of the scene that may be captured under different conditions. The first reference image is used to recover texture; feature points are extracted along the boundaries of both the source and reference regions, while a warping function deforms the reference region to fit inside the source. The second reference is used to recover color by replacing the mean and variance of the texture reference image with those of the color reference. A user study conducted with both modified and original images demonstrates the benefits of our method. The results show that a majority of the enhanced images look natural and are preferred to the originals.