Removing shadows from color images

Date created: 

This thesis is concerned with the derivation of a shadow-free image representation. We propose several methods of automatically removing and detecting shadows in a color image. Our methods stem from the illuminant invariant theory which requires a camera calibration step to find the direction of illumination changes. In our work, instead of a camera calibration we aim at finding this direction from evidence in the color image itself. Specifically, we recognize that producing a 1-d projection in the correct invariant direction which is orthogonal to the direction of changes of illumination, will result in a 1-d distribution of pixel values that have smaller entropy than projecting in the wrong direction. To be able to develop an effective description of the entropy-minimization task, we go over to the quadratic entropy, rather than Shannon's definition. Replacing the observed pixels with a kernel density probability distribution, the quadratic entropy can be written as a very simple formulation, and can be evaluated using the efficient Fast Gauss Transform. The entropy is more insensitive to quantization than is the usual definition. The shadow removal step produces good shadow-free color image results whenever strong shadow edges are present in the image. In almost every case studied, entropy has a strong minimum for the invariant direction, revealing a new property of image formation. Shadow detection per se is an important step in image analysis. We propose a method of detecting not just strong shadow edges but indeed entire shadow regions in the image taken under ambient light. We argue that the difference in a log domain of the flash image and the ambient image gives a very simple feature space consisting of two components - one in an illuminant-change 3-vector direction, and one along the gray axis. This space provides excellent separation of the shadow and nonshadow areas. We propose a method for efficient ambient illuminant estimation using the flash image. We design a novel white balance scheme which uses the white patch under the estimated illuminant as reference white color for balancing images.

The author has placed restrictions on the PDF copy of this thesis. The PDF is not printable nor copyable. If you would like the SFU Library to attempt to contact the author to get permission to print a copy, please email your request to
Document type: 
Copyright remains with the author
School of Computing Science - Simon Fraser University
Thesis type: 
Thesis (Ph.D.)