View Dependent Enhancement of the Dynamic Range of Video.pdf

View Dependent Enhancement of the Dynamic Range of Video Matti Niskanen Machine Vision Group, Infotech Oulu, University of Oulu, Finland matti.niskanen@ee.oulu.fi Abstract frame (d). The lower row in the figure shows some details (with histograms stretched for visualization purposes). The There are many applications for computer vision where a HDR frame preserves details of both dark and light regions. scene observed contains a wide range of brightness. Often, Most of the approaches related deal with still images and the low dynamic range of a camera limits the accuracy of in- contribute mostly to solving unknown camera response func- formation that can be extracted from the video. Frames may tions [1, 2, 8, 9] that are needed in order to produce a realistic contain saturated pixels of bright targets, poor resolution and looking image. We use a CCD camera, where response is noisy data for dark regions, or both. either linear or easily linearized. Also we set the exposures In this paper, we propose a method for generating high dy- automatically ourselves, so we do not need to estimate these namic range (HDR) videos by combining successive frames. from the scratch. Our method is mainly intended as a pre-step The first phase is to set the exposures for each frame con- for vision problems, and we do not try to please the human tributing to one HDR frame. The exposures are automatically eye, nor do we therefore consider tonemapping either. adapted according to image contents to provide a maximum