Modeling shapes using uniform cubic B-splines for rice seed image analysis

A piecewise model for representing and matching rice seed shapes is presented to be used for visual phenotyping of rice accessions. Scanned seed images for an accession are analyzed to characterize its seeds' shape. The resulting shape is modeled using uniform cubic B-splines, and saved to a database of representative splines for each accession. Databases using uniform cubic B-Splines with four, eight, 12, 16, and 20 curve segments were generated, and two classifiers for matching new, unknown shapes to the known accession shapes were tested. Based on tests, using uniform cubic B-Splines with four curve segments yielded promising results, and using eight, 12, 16, or 20 curve segments did not improve the classifiers' accuracy significantly.