Design and Development of Ficus Species Database and 2D Leaf Image Identification System

Plants are important part of the ecosystem in the world. Numerous studies had been done on the richness of plants diversity. There are many plant databases currently available online. A search for useful information regarding a particular plant can be performed easily through the databases using text such as the name of a plant. However, this task will be difficult if only image of the plant is available. Thus, to facilitate the search using image of a leaf, a simple two-dimensional (2-D) Ficus Identification Database System (FicIDS) was developed. This system can perform search using both text information and image of the plant. Basically, FicIDS system was focused on Ficus species. Ficus species were chosen due to its variable leaf shapes and its significance in our local herbal industries. Currently, there is a high demand for the natural products derived from this plant, particularly from Ficus deltoidea. But, it is very difficult to identify a Ficus plant correctly since there are more than 100 different Ficus species and more than 20 different varieties of F. deltoidea available in Malaysia. Furthermore, there is no proper documentation of this plant. The FicIDS system was designed and developed to identify Ficus plants based on the leaf image and to store the data about these species. Herbarium specimens of Ficus plant were prepared as evidence for the plants used in this study. Images of herbarium specimen and live plant materials were captured and stored in the database. Additional text information on Ficus plants were also collected from various sources to build up the database. Microsoft Office Access database management system was used to develop the plant database on Windows XP platform. 2-D leaf images identification system was constructed using the MATLAB R2006a program. The shape and size of plant leaves were used as the main features to identify a particular Ficus species. The process of image identification system comprised of four steps, namely image acquisition, image preprocessing and features extraction, computing of descriptors values and normalization, and k-nearest neighbor classification and decision making based on Euclidean distance. Thirteen descriptors values were used in identification of the image, which include aspect ratio, circularity, area convexity, rectangularity, sphericity, eccentricity, and 7 invariant moments. The accuracy of leaf image identification system was evaluated by using 130 leaf images corresponding to 6 Ficus species and 4 varieties of Ficus deltoidea. The evaluation of overall performance of FicIDS system showed that 120 (92.31%) of the tested leaf images were successfully identified by the system. However, the 2-D FicIDS system has some limitations with respect to image identification. The system requires single intact leaf with white background for identification. These limitations may be overcome by using three-dimensional (3-D) leaf image where more features of leaf such as leaf texture or venation can be included to improve the image recognition performance.

PDF FBSB_2010_14_F.pdf Download (156kB)

Actions (login required)

View Item