![]() ![]() Their success relies heavily on high-quality domain-specific training datasets, which this work tries to address for the first time. These methods attempt to decompose an image into “clean” and “corrupt” by training a neural network. While no learning-based techniques exist for flare removal, they have been developed for other artifact removal applications such as reflection removal, dehazing, rain removal, and denoising. To overcome this data collection challenge, we exploit the physically-realistic model and build a novel pipeline to generate a large and diverse dataset of semi-synthetic flare-corrupted images along with ground-truth flare-free images. But this approach is too labor-intensive to produce the thousands or millions of image pairs usually required to train a neural network.įurthermore, this approach only works when the flare-causing illuminant lies outside of the camera’s field of view (see Fig. With significant effort this can be accomplished by collecting pairs of images taken on a tripod where in one image the photographer manually places an occluder between the illuminant and the camera. 1 IntroductionĬollecting a large number of perfectly-aligned image pairs with and without lens flare would be tedious at best and impossible at worst: the camera and the scene would need to be static (a particularly difficult requirement given that most lens flare occurs outdoors and involve the sun), and one would need some mechanism for “switching” the artifacts on and off without also changing the illumination of the scene. Using the semi-synthetic data generated by this pipeline, we build a neural network to remove lens flare.Įxperiments show that our model generalizes well to real lens flares captured by different devices, and outperforms start-of-the-art methods by 3dB in PSNR. We take a principled approach to explicitly model the optical causes of flare, which leads to a novel semi-synthetic pipeline for generating flare-corrupted images from both empirical and wave-optics-simulated lens flares. Existing software methods make strong assumptions about the artifacts’ geometry or brightness, and thus only handle a small subset of flares. The diversity in its appearance makes flare removal extremely challenging. It is caused by either multiple reflections within the lens or scattering due to scratches or dust on the lens, and may appear in a wide variety of patterns: halos, streaks, color bleeding, haze, etc. Lens flare is a common artifact in photographs occurring when the camera is pointed at a strong light source. ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |