Grain‐size information from the statistical properties of digital images of sediment

ABSTRACT The autocorrelation technique for estimating grain-size from digital images ofsand beds has been extended and validated for use on coarse sand (0AE7 mm)and gravel (up to 20 mm). A number of aspects of the technique have beenexplored and some potential improvements suggested. Autocorrelation is justone suitable statistical method sensitive to the grain-size of sediment in digitalimages; four additional techniques are presented and their relative meritsdiscussed. A collective suite of techniques applicable to the general problem ofgrain-size estimation from digital images of sediment might broaden theapplicability to more sedimentary environments, as well as improve itsaccuracy. These techniques are compared using a large data set from a gravelbarrier beach in southern England. Based on over 180 samples, mean grain-sizeof sieved and imaged sediments correspond to within between 8% and 16%.Some theoretical aspects of the spatial arrangement of image intensity indigital images of natural sediments are addressed, including the fractal natureof sediments in images, which has potential implications for derivation ofgrain-size distributions from images of sand-sized material throughsegmentation and thresholding. The methods outlined in this contributionmay also find application in further uncovering the geometric structure ofsediment beds, as well as in the simulation of sedimentation processes.KeywordsGrain size analysis, digital images, statistical models, gravel beach.INTRODUCTIONGrain-size information from natural environ-ments traditionally is obtained using methodssuch as sieving, laser diffraction and settling,which do not help measure equivalent propertiesof the sample. In addition, the slow and labour-intensive nature of these methodologies has lim-ited the spatial and temporal resolution at whichgrain-size data are obtained. Where grain-size isan important parameter, such as in determiningsediment transport, studies can be limited funda-mentally by the difficulty in sampling for sedi-ment at the required frequency. A furtherdisadvantage is that sediment must be removedfrom the natural environment, potentially alteringsubsequent system development. In contrast,photographic methods can be used to measuresediment size at a resolution comparable withmeasurements of hydraulic, hydrodynamic andmorphological/topographical conditions withoutdisrupting the sediment body by direct sampling.The problem of deriving sediment size infor-mation from digital images of sediment has beenapproached using two different families of tech-niques. The first is based on edge detection andimage segmentation principles (e.g. Butler et al.,2001; Sime & Ferguson, 2003; Graham et al.,2005). Such techniques rely on marked image-intensity contrasts between grains and gapsbetween grains (interstices), such that thresholdscan be specified to discriminate individual grainsfrom the background intensity levels (e.g. Sime &Ferguson, 2003). Thus far, these methods areSedimentology (2009) 56, 421–438