Plant Tissue Slice Image Sequences Based On Three-dimensional Visualization

Plants are known as important research objects in agricultural science. One of the current hottest topics in digital agriculture research is to build digital models of plants and their internal tissue structures based on visualization techniques. This digital model is thought helpful not only to reveal the internal structure-function relationship of plants, but also to promote the interdisciplinary overlapping and fusion of modern agriculture studies. By digitizing plant’s structure and function, an interactive three-dimensional model can be used to describe plant’s apparent and internal structures, and help users conveniently compute the shape, size and spatial relationship of plant internal organization. Therefore, The three dimensional modeling of plant tissues has important practical significance and application prospects to classify and identify features of plant tissues, to assist intuitively and visually teaching of botany, to popularize the relationship of internal structures and physiological functions, and to extend the scientific knowledge of plants.The main contributions in the thesis are summarized as follows:1) Adaptive image segmentation methods of plant tissues. Image segmentation is an indispensable step in the three-dimensional reconstruction of plant tissues. According to the structure features of plant tissues, adaptive segmentation algorithms were developed. The region growing method was used to segment the vessels from cucumber root images. And a new segmentation algorithm based on matching block pattern was presented to extract the vascular bundles of cucumber stems based on their gray distribution features, and verify the accuracy of this algorithm by calculating geometrical factors of vascular bundles, such as axis length and area etc.2) Three-dimensional visualization methods of plant tissues. The Marching Cube, maximum intensity projection and synthetic volume rendering algorithms were respectively used to build and visualize three-dimensional models of vascular bundles of corn internodes, vascular bundles of cucumber stem, and embryo and endosperm of corn seed. Based on the discussion of advantages and disadvantages of these algorithms, the visualization method suitable for serial images of plant tissues was presented. It is the first time to build and visualize three-dimensional models of vascular bundles of corn internodes in China.3) GPU accelerated volume rendering algorithm. Aiming at the large-scale volume dataset of plant tissues, a GPU-accelerated volume rendering algorithm was proposed to improve the render and interaction efficiencies. First, the partitioning strategies for volume data was used to complete a reasonable division of the volume data, and the three-dimensional texture information within the entire block of data was stored in the memory of the GPU. Octree was then used to establish and retrieve texture information of the block. Finally, the empty voxels jump and ray-casting algorithm was implemented to improve the rendering speed with invariant rendering quality.