Automatic estimation of rock particulate size on conveyer belt using image analysis

Image segmentation is an important and difficult step for automatic rock particle size distribution estimation. In this paper, we propose a method for the segmentation rock images in a machine vision system using the Voronoi diagram. Typically, rock edge detection is achieved using the watershed transform. Marker-driven watershed segmentation extracts seeds indicating the presence of rocks at specific image locations. The marker locations are then set to be regional minima within the topological surface, which is normally the gradient or thresholds of the original input image. In contrast, our approach uses the generalized Voronoi diagram through Euclidian distance transform for the segmentation of the rock image after extracting the markers. In order to avoid the difficult step of segmentation, we also investigate an image-classification system for rock particulate size estimation using two-level wavelet decomposition. Experimental results show that using the Voronoi diagram is not only more robust than watershed for rock particulate size estimation but also less computationally complex. The Voronoi diagram and watershed methods are superior to the image classification method