Segmentation of contacting soil particles in images by modified watershed analysis

Image-based soil particle size and shape characterization relies on computer methods to process and analyze the images. For contacting particles spread on a flat surface this requires delineation of particle boundaries through shape-based image segmentation. The traditional method using watershed analysis fails for particles that have constrictions (are peanut-shaped). The oversegmentation interprets such particles as being two, thereby underestimating the long particle dimension by about 50% and overestimating particle sphericity by about a factor of two. This paper presents a solution to the problem of oversegmentation through morphologic reconstruction. The key to this improvement is distinguishing the necks in peanut shaped particles from actual contacts between particles. A parameter a is defined to quantify the necks and contacts. Approximately 220,000 particles in a range of 2.0-35.0 mm having various shapes and angularities were studied to find typical a values for necks and contacts. An algorithm is proposed to correct the oversegmentation based on a. The results show that this improved watershed analysis accurately segments sand particles at contacts while preserving the continuity of peanut shaped particles. Example lab tests demonstrate the significance of the problem and its solution. (C) 2015 Elsevier Ltd. All rights reserved.