Recognition of Plants with Complicated Background by Leaf Features

Abstract

Leaf classification is a significant and meaningful work. However, the interference and overlapping of objects may affect the recognition effect of leaves with complicated background. In this paper, a hybrid framework of classifying leaves with complicated background is proposed. Firstly, a novel watershed segmentation based on iterative opening and closing reconstruction is introduced to segment leaves from complicated background, which contains texture and shape information of leaf. Then, the block Local Binary Pattern(LBP) operators, whose dimensionality is reduced by locally linear embedding(LLE), are extracted as texture features. In addition, the shape features of leaves are described with the Fourier descriptors. Finally, the texture features and shape features are combined as the input of Support Vector Machine(SVM) classifier to realize the accurate classification of plant species. Experimental results show that the proposed method is effective.