Automated Sizing of Coarse-Grained Sediments: Image-Processing Procedures

This is the first in a pair of papers in which we present image-processing-based procedures for the measurement of fluvial gravels. The spatial and temporal resolution of surface grain-size characterization is constrained by the time-consuming and costly nature of traditional measurement techniques. Several groups have developed image-processing-based procedures, but none have demonstrated the transferability of these techniques between sites with different lithological, clast form and textural characteristics. Here we focus on image-processing procedures for identifying and measuring image objects (i.e. grains); the second paper examines the application of such procedures to the measurement of fluvially deposited gravels. Four image-segmentation procedures are developed, each having several internal parameters, giving a total of 416 permutations. These are executed on 39 images from three field sites at which the clasts have contrasting physical properties. The performance of each procedure is evaluated against a sample of manually digitized grains in the same images, by comparing three derived statistics. The results demonstrate that it is relatively straightforward to develop procedures that satisfactorily identify objects in any single image or a set of images with similar sedimentary characteristics. However, the optimal procedure is that which gives consistently good results across sites with dissimilar sedimentary characteristics. We show that neighborhood-based operations are the most powerful, and a morphological bottom-hat transform with a double threshold is optimal. It is demonstrated that its performance approaches that of the procedures giving the best results for individual sites. Overall, it out-performs previously published, or improvements to previously published, methods.