A new binarization method for non-uniform illuminated document images

An original binarization method for document images acquired under non-uniform illumination conditions was proposed in this paper. The Curvelet transform and Otsu's method were combined to binarize the non-uniform illuminated images. The non-uniform illumination image was decomposed by the Curvelet. And, the Curvelet coefficients were enhanced by nonlinear functions. Then, the reconstructed image was processed by Otsu's binarization method. The experiment results of our and the others methods were shown and discussed.

► The images illuminated by five non-uniform patterns.
► The image was represented by Curvelet coefficients.
► The Curvelet coefficients were processed non-linearly.
► The images were binarized by our method and the other classical methods.
► The results, running time and the correct rate of OCR recognition were listed for comparison.