Research On Shape Description And Matching Based On Contour And Skeleton

With the development of the Internet, we can obtain and make use of more and more image information, how to retrieve out those image information we need from these image information has being one of the current more active research fields. Shape features as one of the low-rise features, as one of the important tools for describe target and objects in the image, play a crucial role in content-based image retrieval technology. Considering the skeleton and the contour of shape, this thesis studys the shape description and matching problem, including skeleton graph matching,shape description based on the skeleton and shape description and matching based on contour.In this thesis, the main contents and innovative work is as follows:(1) Considering the skeleton of shape, this thesis proposes a new shape description and matching method based on the tree structure constructed by skeleton. First according to certain algorithm this thesis constructs tree structure by skeleton, then describes skeleton endpoints with information of geodesic path from the root node to leaf nodes, finally uses the time series matching algorithm to determine the relations between the leaf nodes of two skeleton trees, measures the similarity between shapes through matching information. The algorithm has the characteristics of translation,rotation and scale invariance, also can better to handle the changes of connection part of shape, has centain robustness in non-rigid situation and low time complexity.(2) Considering the skeleton of shape, this thesis proposes a new shape description method based on statistical information of skeleton structural features. First defines new skeleton feature, this thesis calculates structure feature information between skeleton points, statistically constructs shape descriptor through structure feature information. This shape descriptor not only has the characteristic of rigid invariance, but also can achieves higher efficiency and retrieval precision in non-rigid situation. (3) Considering the contour of shape, based on complex network theory this thesis proposes a new shape description and matching method. First of all, based on inner distance of shape, a shape is modeled into a small-world complex network, then this thesis analyses multiscale complex network model, constructs shape descriptor based on multiscale histogram, and finally achieves shape matching with description based on complex network. Experimental results show that the proposed algorithm can achieve shape matching with non-rigid transformations and the higher precision in shape retrieval.