Feature extraction of overlapping hevea leaves: a comparative study

Abstract

Automation of rubber tree clone classification has inspired research into new methods of leaf feature extraction. In current practice, rubber clone inspectors has been using several leaf features to identify clone types. One of the unique features of rubber tree leaf is palmate leaflets. This characteristic generates different leaflet positions, where the leaves are overlapping or separated. In this research, we propose keypoint extraction and line detection methods to extract shape and axil (angle between petioles) features of leaflet positions. The results of keypoint extraction methods, namely, SIFT, Harris, and FAST, were compared and discussed for shape feature extraction. Next, Hough transformation and boundary-tracing methods were compared to identify the suitable axil detection method. The evaluation result demonstrates the proper keypoint extraction method for shape context and the clear advantages of Hough Transformation in accuracy of angle detection.