A Leaf Image Retrieval Scheme Based on Partial Dynamic Time Warping and Two-Level Filtering

Abstract

In this paper, we propose an efficient leaf image retrieval scheme based on its shape feature. To represent leaf shape, we first derive a distance curve from each leaf using distances between its center and the points along the contour. Next, for the storage and matching purpose, we extract a set of Fourier Coefficients (FCs) from the curve. Typically, a set of FCs fails to restore the original curve completely. In order to reduce the difference between original and restored curves, we divide whole curve into several partitions such that each partition has a small difference with the original curve. For image matching, we propose a Partial Dynamic Time Warping (PDTW) algorithm. We also propose a categorization scheme using the number of unit curves for filtering purpose. In the experiment, we show that our proposed scheme improved both efficiency and accuracy of leaf image retrieval.