Images thresholding via within-cluster weighting variance
Shuhong Yang, Tong Zhang · Proceedings of the 2021 5th International Conference on Electronic Information Technology and Computer Engineering · 2021
As one of the most popular thresholding methods, Otsu method is known to produce biased threshold when variances among data classes are significantly different, yet most researches point out this deviation mainly by experiments. In this paper, the essential cause of the deviation of Otsu algorithm is investigated theoretically, then starting from the analysis of two improved versions of Otsu algorithms, a thresholding method based on within cluster weighting variance is proposed, in which two parameters are optimized to adapt to images with all kind of distribution. Experimental results on both synthetic and real images demonstrated that the proposed method is effective and robust. Furthermore, the proposed method is especially suitable for segmenting serialized images which is common in real application.