Texture-Consistent Shadow Removal

Feng Liu and Michael Gleicher

Computer Sciences Department, University of Wisconsin-Madison


This paper presents an approach to shadow removal that preserves texture consistency between the original shadow and lit area. Illumination reduction in the shadow area not only darkens that area, but also changes the texture characteristics there. We achieve texture-consistent shadow removal by constructing a gradient field that removes these two shadow effects and reconstructing from this gradient field a shadow-free image by solving a Poisson equation. The key is to construct the shadow-free and texture-consistent gradient field. First, we estimate the illumination change surface and remove the gradients it induces. We estimate and sample the illumination change surface using line segments across the shadow boundary. We model the illumination change along each line segment as a spline. Because the illumination change surface is smooth along the shadow boundary, we formulate estimating these illumination change splines along the boundary as an optimization problem which balances the smoothness between the splines and their fitness to the image data. Using these illumination change splines, we cancel the illumination change effect on the original gradient field. Second, we estimate the shadow effect on the texture characteristics using the texture in the umbra and lit area near the shadow boundary. Accordingly we transform the gradients inside the shadow area to cancel this shadow effect to be compatible with the lit area. Experiments on real photos from Flickr demonstrate the effectiveness of our method.

(a) original image

(b) shadow boundary specified by user

(c) shadow boundary specified by user (zoom in)

(d) shadow- free image

Shadow removal example. Given a rough shadow boundary P provided by users (b and c), our algorithm removes the shadow (d). The red curve inside the brush stroke is the trajectory of the brush center. Users do not need to provide a precise shadow boundary as shown in (c) (Notice the eagle's right wing.). The brush strokes divide the image into 3 areas: definite umbra areas, U, definite lit areas, L, and boundary, P, which contains the penumbra area as well as parts of the umbra and lit area.
Copyright of the above source image belongs to 
Feng Liu and Michael Gleicher. Texture-Consistent Shadow Removal. ECCV 2008, Marseille, France, October 2008. pp. 437-450. PDF