We propose a new method for interactive image color replacement that creates smooth and naturally looking results with minimal user interaction. Our system expects as input a source image and rawly scribbled target color values and generates high quality results in interactive rates. To achieve this goal we introduce an algorithm that preserves pairwise distances of the signatures in the original image and simultaneously maps the color to the user defined target values. We propose efficient sub-sampling in order to reduce the computational load and adapt semi-supervised locally linear embedding to optimize the constraints in one objective function. We show the application of the algorithm on typical photographs and compare the results to other color replacement methods. © 2012 Springer-Verlag Berlin Heidelberg.
Bibliographical noteKAUST Repository Item: Exported on 2020-10-01
Acknowledgements: This research was financially supported by Science Foundation Arizona, US Navy, and NSF. We would like tom thank Tom Ang (Fig. 14) and Norman Koren (Figs. 1, 5) for the permission to use their outstanding photographs.