Morphology and Texture Features Extraction for Leaf Sheet Image Retrieval.

Abstract

In the biology and forestry science, there are many kinds of leaves with different features. It causes difficulty to make leaf identification. Colour feature is not too significantly different because generally they are green. So, the primary leaf identification depends on morphology and texture feature. The two of them makes leaf identification more accurate. But, the combination makes some varies. A lot of people even the scientist are difficult to recognize the leaf, especially in leaf image. The reason to make feature extraction automatically by computer will be used for leaf content based image retrieval. This research started with collected leaf images by taking some pictures of them with digital camera. The collected location was Bogor Agricultural University Darmaga area. Eleven kinds of fruit tree leaves were photoing, one hundred pictures for each kind. So, there are eleven classes of leaf on database. The next step was cleaning the images. Background images were cleaned to get white background. After that, Digital Morphological Features Extraction was used to extract the shape features and Cooccurrence Matrix for texture features. Three primary features of shape are extracted and calculated it to get three derivative features. Meanwhile, seven features of texture are extracted with Coocurrence Matrix. Those features were combined with Bayesian Network Model to improve content based image retrieval. Generally, Bayesian Network Model increases precision value of image retrieval. Average precision value for Bayesian Network Model on all class is 0.2733. This is the highest value. Average precision value using texture feature on all class is 0.2722. Then, average precision value using shape feature on all classes is 0.2531. This is the lowest value. Keyword : Content Based Image Retrieval (CBIR), feature extraction, Morphological Features Extraction, Co-occurrence Matrix, Bayesian Netrwork Model.