Mark Grundland
University of Cambridge
ABSTRACT:
In image and video editing, image compositing is the process of combining pictures, commonly know as photomontage. A key technical challenge is to produce composites that preserve the visual quality of the components used to create them. Linear interpolation is the standard method of image blending used in image compositing. As an averaging operation, linear interpolation acts to reduce variation. As a result, its output can look less appealing than its input. Composite images often leave something to be desired. When creating layered visual effects, compositing artists may need to manually compensate for faded contrast, dull colors, lost details, or lack of emphasis. To address these long standing shortcomings of image blending, we redefine the fundamental operation of linear interpolation, reformulating its weights, its operators, and its results.
We aim to give users visually meaningful, high level control over the appearance of composite images. We propose four novel image blending operators that are designed to maintain key visual characteristics of their inputs: contrast, color, detail, and salience. Our contrast preserving technique applies a linear color mapping to recover the contrast lost due to linear interpolation. Our color preserving technique extends homomorphic image processing by establishing an isomorphism between the image colors and the real numbers, allowing any mathematical operation defined on real numbers to be applied to colors without losing its algebraic properties or mapping colors out of gamut. Our detail preserving technique relies on the signed weighted power mean to emphasize variation over uniformity combining image features at different levels of detail. Finally, our salience preserving technique retains the most informative regions of the input images by balancing their relative opacity with their relative saliency. Applying our approach, casual users may benefit from image compositing tools that do not require highly accurate image mattes to create compelling composites, while expert users may take advantage of contrast enhancement and color correction tools to ensure sharp contrast and vivid colors.
This seminar presents the work of Mark Grundland, Rahul Vohra,
Gareth P. Williams, and Neil A. Dodgson at the University of
Cambridge, which was recently published as "Cross Dissolve Without
Cross Fade: Preserving Contrast, Color and Salience in Image
Compositing" at EUROGRAPHICS and "Nonlinear Multiresolution Image
Blending" at ICCVG.
CONTACT: Mark Grundland (mark@eyemaginary.com)
PROJECT: http://www.eyemaginary.com/compositing/compositing.html