Rock particle images are different and complex, which is depending on various situations, in most cases, traditional image segmentation methods often make under-segmentation and over-segmentation. In order to overcome the disadvantages, a new algorithm based on improved graph based algorithm (Normalized Cut) is proposed. In the algorithm, to denoise and reduce data size, an image is reduced by scale transformation; and image segmentation is made by an improved Normalized Cut algorithm. In

more »

... ments, different types of rock particle images were tested, and the new algorithm was compared with the traditional algorithms such as Threshold, Edge detection, Minimum spanning tree, Clustering analysis, Fuzzy C-Means and Watershed etc. The testing results show that the image segmentation effect of the new algorithm is better than that by the traditional image segmentation algorithms. The studied algorithm was applied into the rock blasting in a large open-pit mine, and the results were satisfactory. The clustering optimization [16] based on graph theory was studied in the 1960s firstly, and then, it was gradually applied in the field of image processing and analysis in the early 1980s. The image segmentation method based on graph theory is a top-down global segmentation method, which is robustness [12] , and it combines the advantages of the graph method that is easy to process the local data characteristics and the advantage of objective function. The main idea of graph theory [12] [13] [14] is to map an image into a weighted undirected graph by taking each pixel as a node in graph, and to connect each pair of nodes by a graph edge. The weight on the graph edge corresponds to the neighboring relations between two pixels in the processed image. In the graph procedure, the image segmentation is completed by using various segmentation criteria. In the computing, a graph is regarded as a matrix, where each pixel corresponds to each element in the matrix. According to the value and position relationship of each element, a required classification can be got by using mathematical calculation. Figure 1 shows the mapping relation between an image and a graph. information reflects on the particle surfaces, with less under-segmentation problems; and Clustering analysis based algorithm is also not suitable for this kind of rock particle images, the segmentation effect is better than that in (d), just having some under-segmentation problems in the regions of small particles; Watershed algorithm operated in images is similar to that in (c), some problems of under-segmentation and over-segmentation still exist. The algorithm in this paper can solve the over-segmentation problem, the main particles are extracted, and the under-segmentation is less by comparing with the other algorithms.