Automated extraction of grain-size data from gravel surfaces using digital image processing

This paper describes and tests a method for the automated extraction of grain-size data from digital imagery. It combines two basic image processing methods for this purpose: grey-scale thresholding to create a binary image and watershed segmentation to grow edges on the binary image to allow the identification of individual grains. The method is subject to rigorous testing in terms of edge detection and automatic measurement of grain-size information from the edge images, and is also compared with the results obtained from simple direct clast sampling. The edge detection methods are tested with respect to manually-identified edges. This suggests that simple thresholding of raw imagery produces grain-size estimates that are: (i) in excellent agreement with manual estimates, above a critical particle size defined by the scale of the photography; (ii) downgraded with the inclusion of additional edge information from analysis of high resolution digital elevation models (DEMs); and (iii) not affected by the use of raw imagery as opposed to imagery that has been rectified to deal with geometric, tilt and relief distortion effects. The automated ellipse-based measurement method is shown to produce a good estimate of two-dimensional a- and b- axes as they appear as long and short axes on the edge images. Thus, the research shows that it can be used to map and quantify very rapidly spatial variations in grain-size characteristics, although it cannot deal with the long-recognised problem of the relationship between two-dimensional planform grain-size estimates and actual a- and b- axes obtained by direct grain sampling.