An unsupervised method for flotation froth image segmentation evaluation base on image gray-level distribution

In the machine vision based flotation process monitoring and control, it well accepted that the accurate froth image segmentation acts as the foundation of morphological features extraction of flotation bubbles, whose important roles of closely relating to the flotation production states have long been acknowledged. However, automatic and unsupervised froth image segmentation evaluation far has been largely subjective due to the complexities and specificities of the froth images. Namely, the conventional froth image segmentation quality is mainly judged by only on intuition, which cannot make sure that the morphological features of the bubbles measured by the image segmentation algorithms in the vision monitoring system are always accurate and representative of the real flotation production states. To advance the machine vision based flotation process monitoring and control for accurate morphological feature extraction, a novel unsupervised froth image segmentation evaluation method based on image gray-level distribution according to the characteristics of froth images is proposed, since the ground truth of froth image segmentation is always difficult to obtain. The number of highlights in the froth image is extracted and used as the subjective evaluation index by analyzing the gray-level distribution of each individual segmented region. The experimental results demonstrate that this method can offer effective qualitative evaluation results of the flotation froth image segmentations without providing the ground-truth boundaries of the froth images, which ultimately leads to the adaptive froth image segmentation algorithm selection and proper algorithm parameters setting to achieve the accurate morphologic feature parameters of the froth bubbles.