Minimum Class Variance Thresholding Based on Multi-objective Optimization

Variance-based thresholding is one of the most popular methods for image segmentation. The mechanism of variance-based thresholding methods is to minimize the class variance. A novel minimum class variance thresholding method based on multi-objective optimization has been presented, and the ideal threshold is achieved by minimizing the variance of each class and the sum of them, and this will lead to more satisfactory segmentation result. The presented method possesses the merits of restraining the class probability and the class variance effects, and it is more accurate. Firstly, the proposed method is compared quantitatively with other methods on lots of synthetic images with the convenience of obtaining the ideal thresholds precisely and the ground-truth images exactly. The presented method possess better performance at most magnitudes of the noise. At the same time, experiments over real infrared images and visual images also have illustrated the better performance of the presented method.