This paper presents an algorithm that enhances undesirably illuminated imagesby generating and fusing multi-level illuminations from a single image.Theinput image is first decomposed into illumination and reflectance components byusing an edge-preserving smoothing filter. Then the reflectance component isscaled up to improve the image details in bright areas. The illuminationcomponent is scaled up and down to generate several illumination images thatcorrespond to certain camera exposure values different from the original. Thevirtual multi-exposure illuminations are blended into an enhanced illumination,where we also propose a method to generate appropriate weight maps for the tonefusion. Finally, an enhanced image is obtained by multiplying the equalizedillumination and enhanced reflectance. Experiments show that the proposedalgorithm produces visually pleasing output and also yields comparableobjective results to the conventional enhancement methods, while requiringmodest computational loads.
展开▼
