Fragment size estimation without image segmentation

Abstract

When rock fragment particles are densely packed on a moving conveyor belt, it is difficult to use an image segmentation algorithm to delineate every particle online, because fragment particles are of different colours, 3D geometry properties, and overlapping to each other. Given image sequences of closely packed fragment particles, the underlying aim is to estimate fragment size directly (without using an image segmentation algorithm). In a way, this is similar to the task of counting objects without directly counting them. Such calculations may, for example, be useful for fast estimation of particle size in different application areas. The topic is that of estimating average size (=average diameter) of densely packed fragment particles online, from formulas involving edge density. The authors use, primarily, edges from moment based thresholding. An average shape factor is involved in the calculations, obtained from some frames on crude partial segmentation. Measurement results from several hundreds of images have been analysed.