Matting and Compositing in Radiance Space

Digital matting is one of the most important techniques in the film and video industry. The illustration model of the foreground object is extracted from photographs or videos and can be combined with new background image. Almost all of the existing digital matting and compositing algorithms are computed in RGB color space. However, because of the nonlinearity of the imaging system, the relationship between the color value, sampled by the imaging device, and the correspondent radiance value is approximately logarithmic. The matting equation described in color space leads to an unfaithful physical model of the matting problem. It makes errors in extracting and compositing the matte data of reflection or refraction surfaces. A matting and compositing method in radiance space is presented in this paper. With recovering the Hurter-Driffield curve of the imaging system, we introduce the matting problem into radiance space. We compare the results of triangulation matting and compositing in color and radiance space. It comes out that the matting data can be calculated more accurately in radiance space