Research On Shape Based Leaf Image Retrieval And Clustering

Recent years have witnessed the rapid growth of digital images due to the increasing power of computing and the fast development of internet. The development of powerful retrieval tools has become a central problem in various machine vision applications. A new multimedia application, called content based image retrieval (CBIR) has come into being to address this urgent issue. In CBIR, image is described by several low level image features, such as color, texture, shape or the combination of these features. Shape is an important feature that considers the global feature of object. Applications of shape analysis include biomedical image analysis, morphometry, database retrieval, military target recognition and computer vision. The focus of this thesis is on the field of shape based leaf images retrieval and clustering, their effectiveness are evaluated and validated via some experiments.One goal of this thesis is to present visual descriptors that characterize the leaf shapes. Numerous shape representation and description techniques have been reviewed in the thesis. The majority of the methods presented consider the shape using Fourier description of the boundary line of the object. For this kind of shape description, a number of contour-based shape representation and description techniques are studied using a standard methodology,including convergence study, retrieval effectiveness and comparison. After that, a centroid distance histograms method is proposed to represent the shape, this similarity of two shapes are judged by comparing the number of distances in corresponding ranges. Retrieval results show that using combined features outperforms than a single feature.Another goal is to present tools that cluster images according to the shapes of their boundaries, the purpose is to improve database searches in systems with shape-based queries. First, classical clustering algorithms on Euclidean spaces are reviewed; they are well-researched and generally fall into two main categories:...