"Physics-based colour models in computer vision, colour science, graphics and multimedia"

点击翻译按钮获取中文摘要

A.) Computational Photography: (i) I have recently generated re-integrated images starting with an image's desired gradient field pair, a longstanding problem, with greatly reduce artifacts. (ii) In a 2011 patent I show how to transfer the contrast from hi-D image data down to lower dimensional images, e.g. transfer the contrast from RGB+infrared down to colour images. This work uses the patent in (i); however, the solution found is global over the image. A local mapping should in principle be much more effective, and I propose a variational-equation solution. (iii) Then other problems could be attacked, such as image sharpening without halo artifacts, and anisotropic smoothing ("wrinkle remover"). B.) Graphics/Visualization: (i) In work on multiple-lighting image sets, lighting directions are required. But for "in the wild" objects directions are inaccurate, so instead I wish to discover them, along with finding lights' visibility subspaces. (ii) The above could be used to discover damage to artworks lent out by museums. (iii) In hi-D medical image visualization, I have generated perceptually meaningful colour outputs. However, clipping often occurs. If gamut-fitting is included into the optimization this problem could be solved whilst generating the visualization. C.) Computer Vision: (i) Shadow-free intrinsic (illumination invariant) images utilize the entropy in a feature space as a crucial step; I propose using video information, accumulating evidence over time. (ii) Another direction would apply the intrinsic image feature to dermoscopy images: the intrinsic image can be used to eliminate light-colour and intensity changes. I also propose removing skin colour, for a colour-image lesion atlas independent of skin colour. D.) Multimedia: If actors in large theatres were crowd-tracked in 2-D, audio depth plus a panning could be generated. E.) Colour: (i) A substantial proportion of papers use colour incorrectly. I propose a publication and reference codeset for a hierarchy of camera calibrations, from complete to simple, for guiding researchers in how to proceed. (ii) Colour Balance is critical for camera data. I recently presented a new algorithm to the iPhone Camera team at Apple and propose further investigating this.