BINARY IMAGE SEGMENTATION OF AGGREGATES BASED ON POLYGONAL APPROXIMATION AND CLASSIFICATION OF CONCAVITIES

摘要

The aim of image analysis of aggregates for mining and mineral industry is to evaluate the quality of natural and crushed aggregates. In this paper, a new heuristic search algorithm for touching aggregates in a binary image is presented. The algorithm first applies a polygonal approximation for every object where several particles may touch each other; secondly classifies concave points on object boundaries into different classes based on its angle and lengths of 2-vertex lines; thirdly finds the candidates of start and end points where the end point is not necessarily a concave point; then uses a supplementary cost function to determine if a split path can be accepted, in which variables such as the shortest distance, the shortest relative distance, minimum number of unmatched concave points, “opposite direction” (see text), particle area and maximum ratio between two split parts in terms of areas, are applied. The algorithm can split not only simply touching particles (two or three particles touching each other), but also large clusters of particles. The algorithm also includes one routine to treat the situation of having one or more holes inside an object. The algorithm has been coded and tested in an on-line system for measuring crushed aggregates in a gravitational flow. In addition to this, the algorithm has also been tested for other different particle images in which particles touch in a complicated fashion. Polygonal approximation and classification of concavities, based on polygons, substantially enhanced the robustness of the algorithm. Test results show that the algorithm works in a promising way.