Hierarchical Shape Decomposition Based On Mathematical Morphology

With the dramatic increase in image resources, object recognition has become the important issue in the image processing. The thesis originates from the project named object recognition in the multi-source image. Object recognition is closely related to object representation. With the in-depth study of various object representations, selecting the appropriate feature space is critical. The shape of the object contains a lot of visual information, becoming the most commonly used feature space. With the in-depth study, it is found that skeleton is widely used in object recognition. Extract skeleton, decompose it and organize segments into a graph. The graph contains all the information and lay foundation for the use of graph matching technology in object recognition. Meanwhile, set the threshold of significance to reduce the impact of noise on the skeleton.In the thesis, shape decomposition is selected as the research object. Because a lot of studies in cognition and psychology have shown that the human visual system tends to decompose the complex object into several simple parts, and then analyze the relationship between them. First study and compare different shape represention methods, focusing on the methods based on mathematical morphology. Implement the morphological skeleton transform and morphological shape decomposition through experiment, and then analyze the main problems.The paper focuses on shape decomposition. Through the experiment, analyze the deficiencies existing in the original algorithm, and through additional definition, concave judgement, gap amendment, area comparision, improve the effectiveness and efficiency. Final analyze the hierarchical relationship of the results.