Implicit 3D Approach to Image Generation: Object-Based Visual Effects by Linear Processing of Multiple Differently Focused Images

Abstract

A new image generation scheme is introduced. The scheme linearly fuses multiple images, which are differently focused, into a new image in which objcets in the scene is applied arbitrary linear processing such as focus(blurring), enhancement, extraction, shifting etc,. The novelty of the work is that it does not require any segmentation to produce visual effects on objects in the scene. It typically uses two images for the scene: in one of them, the foreground is in focus and the background is out of focus, in the other image, vice versa. A linear imaging model is introduced, based on which an identity equation is derived between the original images and the desired image in which the object in the scene is selectively visually manipulated, and the desired image is directly produced from the original images. A linear filter is derived based on the principle. The two original images which are applied linear filters are added and result in the desired image. Various visual effects are examined such as focus manipulation, motion blur, enhancement, extraction, shifting etc,. A special camera is also introduced, by which synchronized three differently focused video can be captured, and dynamic scene can also handled by the scheme. Realtime implementation using the special camera for processing moving scenes is described, too.