A Research On Dynamic Environment Matting

Matting technique includes two processes: matting and compositing. In the matting process, the opacity of the foreground element is extracted from a background image. In the compositing process, the foreground element is combined with a new background image. The environment matte doesn't only describe the transmission of the light, but also the refraction and reflection and so on. For dynamic environment matte, when the view point is moving, the matte is changing as well. Therefore, the analog of dynamic environment matting is much difficult. We improve the existed algorithm of the matte and develop an algorithm for dynamic environment matting in this paper. The main work includes:(1) We develop a method of dynamic environment matting in radiance space. A matting and compositing algorithm which can capture the matte real-time is improved. Currently, the matting and compositing are processed in color space. However, because of a nonlinear map, the color value recorded by the CCD can not describe the radiance of the object truly. To solve this problem, we first get the high dynamic range curve and convert the color space into the radiance space. This approach could reduce blue spill and errors when finding the corresponding region.(2) We study the morphing view developed by Seitz and improve the algorithm: Edge detecting is added before pre-processing to help users figure out corresponding points between two original images fast and accurately. The result is smoother and more natural.(3) A dynamic environment matting algorithm based on image is developed. Currently, the dynamic environment matting algorithm needs to get 3D data of the camera and the foreground object. Therefore, calibration is also needed and the process is very fussy. Addressed the limitation of the algorithm before, the view morphing approach is combined with environment matting and the matting data is interpolated...