The variance entropy multi-level thresholding method

This paper proposes a new multi-level entropy-based image thresholding method. The key principle of the proposed method depends on the minimum of the variance entropy. The method is fully automated at all stages of implementation. It produces competitive segmentation results as compared to the generalized Otsu's method, which is one of the most powerful multi-level thresholding techniques that requires human intervention. In addition, the method significantly outperforms the generalized Kapur's method, which is one of the benchmarking entropy-based thresholding techniques. The method is successfully applied to several scenarios of trial histograms and real images, and its performance is checked using a variety of classification measures and quality metrics.

Automated summary

This paper proposes a new multi-level entropy-based image thresholding method that relies on the minimum of the variance entropy. The method is fully automated and produces segmentation results comparable to the generalized Otsu's method, which requires human intervention. It also outperforms the generalized Kapur's method in benchmarking entropy-based thresholding techniques. The method is successfully applied to various scenarios and its performance is checked using classification measures and quality metrics.