Illumination Correction for Stitching Images

Sascha Klement, Fabian Timm, and Erhardt Barth

Here, we provide additional material concerning our submission "Illumination Correction for Stitching Images" for the ICPR 2010. The following table shows input images with different textures under varying lighting conditions and the results after illumination correction with the the three studied boundary conditions.

Remarks:

• All input images were captured using a Baumer TXG14c camera (1392 x 1040 pixels)
• We focused on the reduction of boundary artifacts when removing illumination inhomogeneities, not on a perfect stitching of the corrected images. So for tiling, the images were simply placed in a 3-by-3 matrix without any further processing. Thus, repetitive patterns are obvious even with polynomial regression, but they are not caused by illumination inhomogeneities.
• With the replicate boundary condition dark areas at the image transitions are clearly visible. Linear extrapolation reduces these boundary artifact significantly. Polynomial Least-Squares Regression gives no further visible improvements but is mathematically the more stable method.
  
• To get the full-resolution image, click on an image.
  

Input Image	Corrected and Tiled Image (Replicate Boundary)	Corrected and Tiled Image (Linear Extrapolation)	Corrected and Tiled Image (Polynomial Least-Squares Regression)