DIGITAL LOG OF THE MAPLETON, UTAH MEGA-TRENCH: OPTICAL ANALYSIS OF A TRENCH WALL USING MORPHOLOGICAL IMAGE PROCESSING OPERATIONS
The clast-segmentation algorithm consists of an optimally, linked sequence of morphological image processing techniques: histogram normalization, thresholding, edge detection, edge linking, watershed transform, opening, and dilation. The algorithm enhances contrast between clasts, eliminates matrix, and labels clasts as individual, watershed regions for which eccentricity, area, perimeter, axes lengths and orientation can be calculated.
To create the digital log, a composite digital photo of the trench wall was subdivided into two hundred 400 x 400 pixel images. The algorithm was autonomously applied to each image resulting in segmentation of over 100,000 clasts. Initial accuracy varied between 70% and 95%, depending on contrast in the original digital photo. Accuracy was improved prior to final labeling and statistical analysis by minimal user-controlled reclassification. Preliminary statistical comparisons indicate three measures (clast-to-matrix ratio, clast eccentricity, and clast orientation) potentially may be used to statistically differentiate colluvial wedges, debris flows and channel deposits exposed in a trench wall. Future work will determine whether these measures can definitively be used to categorize trench wall stratigraphy.
The proposed algorithm is not restricted to clast segmentation and analysis in trench walls. Other applications include statistically analyzing landslide surfaces, stream braiding patterns, down-borehole images, joint patterns, distribution of ground shaking, and Mars or other planetary surfaces.