Computer-implemented digital video object compositing

Abstract

A system and method are disclosed for digitally compositing an object from an input image onto a destination image. The object is composited from an image having an arbitrary or non-uniform colored background containing some non-static elements onto a destination image with reduced effects from shadows cast by the object and with reduced gaps or holes within the object. Various improvements in the compositing procedure such as shadow reduction and hole filling, and less restrictive requirements regarding the object's surroundings are disclosed. A background model is created and a frame of an input image containing the object is obtained. An alpha image is created in which each pixel is either a zero, indicating it is not part of the object, or a one, indicating that it is part of the object. The effect of shadows emanating from the object is reduced so that the composited object in the destination image contains only the object clearly outlined by the object's physical boundaries without the effect of shadows cast by the object. A set of templates is then derived in which each template fits entirely within a designated part of the object. The templates allow holes or gaps in the object created during the compositing process to be filled to a large extent. The object is blended onto a destination image using the alpha image as a blending coefficient.