Project Title: Content-Based Image Retrieval of Herbarium Samples

Activities: We have photographed plant samples from the OSU Herbarium and collected and scanned samples of the corresponding species from trees in Oregon. We have implemented and experimented with algorithms for matching shapes. First, we studied dynamic time warping (edit distance) algorithms for comparing the shapes of two isolated leaves. Then we considered matching isolated leaves against images in which leaves overlap or are incomplete. We imported gap-penalty dynamic programming algorithms from computational biology and implemented and tuned them. We also conducted experiments with using edit distances as approximate kernels in support vector machines. We also studied ways of combining isolated leaf data with data from Herbarium samples to obtain higher classification accuracy.