A. El-ghazal, O. Basir (Canada), and S. Belkasim (USA)
Image segmentation, automatic thresholding, wavelet transform
Image thresholding is concerned with segmenting one or several objects from their background. Finding a threshold value that separates the objects from background with minimum error remains a challenge. A large volume of techniques has been reported to tackle this important issue. It is however recognized that existing thresholding techniques lack acceptable performance in images that are not homogenously illuminated. In this paper, a novel optimal thresholding algorithm is proposed for dealing with non-uniform illumination. The optimal thresholding process is formulated as a problem of optimization that can be solved to determine a threshold value, which maximizes a correlation function between the wavelet-transform coefficients of the original gray level image and its thresholded counterpart. Experimental results are reported to compare the proposed algorithm with commonly used thresholding algorithms. The proposed algorithm demonstrates superior performance in images with uniform as well as non-uniform illuminations.
Important Links:
Go Back
