High dynamic range for dynamic scenes.
Masters thesis, UNSPECIFIED.
Digital cameras, as well as film cameras, are designed to record light so that it can be displayed on computer screen or photographic paper. Unfortunately this limitation prevents the normal cameras to capture the dynamic range of colours (ratio between dark and bright regions) as it is presented in the real world. High Dynamic Range (HDR) photography overcomes this limitation by using a bracketed exposure sequence of the same scene to reproduce a single image, allowing in this way the reproduction of the compressed dynamic range of colors as they are presented in the real world. Unfortunately HDR imagery is not suitable for dynamic scenes. In fact moving objects in the scenes produce in the final HDR images undesirable artefacts called ghosts. For the same reasons a perfect alignment of each image in the sequence is strictly required to obtain consistent and flawless HDR images. In this work, we propose different techniques to adapt HDR imaging to dynamic scenes. These techniques are responsible to detect moving objects in a scene described by a bracketed exposure sequence and to erase the ghosts generated by these movements in the corresponding HDR images. We introduce several techniques based on pixel intensity variance, pixel median value difference and background subtraction. We also tested and defined the best algorithm overall which we identified as the background subtraction based algorithm called FRABS. Besides Movement Detection we propose also two different techniques to align a set of images taken with different exposure times. The first method is based on SIFT feature descriptors, while the second technique is an extension of the Median Threshold Bitmap algorithm originally proposed by Reinhard, Ward et al. Finally we suggest a way to integrate the proposed alignment algorithms and the motion detection techniques into the original Exposure Fusion framework, a novel technique to generate low dynamic range, HDR-like images and proposed by Mertens, Kautz and Van Reeth.
|Title:||High dynamic range for dynamic scenes|
|Event:||University College London|
|Keywords:||Camera Alignment, SIFT, RANSAC, Motion Detection, Exposure Fusion, Background Subtraction, Median Threshold Bitmap, HDR|
|UCL classification:||UCL > School of BEAMS
UCL > School of BEAMS > Faculty of Engineering Science
Archive Staff Only