Pooling-Based Quantitative Approach to Evaluating Binarization Algorithms
Maofu Liu, Ya Liu, Zhenguang Liu, Huijun Hu, Wei Fang · IEEE Multimedia · 2017
To quantitatively evaluate an image binarization algorithm when the ground-truth binary images are unavailable, the authors propose a quantitative evaluation approach based on pooling. First, the binarized images of a sample image from the image dataset (binarized by 27 commonly used binarization algorithms) are put into a pool. Then the pseudo ground truth--two weight matrices of foreground and background points--is produced using the 27 binarized images. Finally, for the image binarization algorithm to be evaluated, a quantitative evaluation operator is generated from the two weight matrices in the pool. The authors ran experiments using two image datasets, and the results show that this quantitative evaluation approach is effective and practical.