A new feature extraction algorithm for overlapping leaves of rubber tree

• Main Author: Anjomshoae, Sule
• Format: Thesis
• Language: English
• Published: 2014
• Subjects: TK7885-7895 Computer engineer. Computer hardware
• Online Access: http://eprints.utm.my/id/eprint/48000/25/SuleAnjomshoaeMFC2014.pdf; http://eprints.utm.my/id/eprint/48000/

Rubber is one of the major sources of national income in Malaysia. Malaysian Rubber Board (MRB) is responsible for the monitoring the quality of rubber to maintain a successful rubber clone breeding program. One of the important factors that affect the quality of raw rubber is the clonal origin of the rubber tree. Currently, clone inspectors classify the rubber tree clones manually using leaf features. There are several features such as leaf tip, leaf base to identify the type of clone. An automated clone classification process is needed to facilitate the inspection process. This research focuses on extracting one of the features for identifying clones which are overlapping leaf features. The challenge of overlapping leaf identification is the similarity of the intensity levels. However, it can be classified using shape and angle of leaves. Therefore, a new feature extraction framework is required to extract shape and angle features. In the new framework, key point extraction method is combined with the nearest neighbor algorithm to extract shape feature. While, angle feature is developed using Hough transform. The proposed method able to detect edge, ridge, and blob features and identify angle between petioles of overlapping leaves. This study identified that angle degrees of the overlapping leaves are in the range between 30° and 55° while angle degrees of non-overlapping leaves are in the range between 55° and 90°. In order to validate the result, the identification method has been tested with fifty rubber leaf images that comprise of both overlapping and non-overlapping features images. The results indicated that forty six overlapping and non-overlapping leaf images matched successfully with correct templates. As a conclusion, the proposed features and their extraction method can be used to identify overlapping and non-overlapping rubber tree leaves.