Image Retrieval Based On Salient Closed Boundary

With the rapid development of internet and multimedia technology, image information has penetrated every industry and become the main form of digital information. As the augmentation of image databases, how to manage and retrieve images is pressing for solution and it's also the research hotspot of international scholars. Technologies of image retrieval have become a new research domain under such circumstance and content-based image retrieval has been the main research direction.Content-based image retrieval consists of image preprocessing, feature extraction, index mechanism and so on. The research of this thesis focuses on feature extraction and index mechanism. A new image retrieval algorithm based on salient closed boundary is proposed in this thesis. This algorithm uses the classic Canny operator to detect edge fragments. Then Minimum Ratio Contour Algorithm in Graph Theory is used to extract the salient closed boundary from edge fragments. At last a new VA-File index mechanism is used to establish images database and histogram intersection method is used as similarity measurement in image retrieval.The main works and innovations of this thesis are as follows:A boundary significance measurement function, which combines boundary and region information, is proposed to solve the noise sensitive problem in salient closed boundary extraction. The effect of this function is tested by contrast experiment.The deficiency of retrieval efficiency based on VA-File is analyzed and an improved scheme of VA-FILE is proposed. The principles, structures, dynamic construction and the implementation of the retrieval algorithm of the improved scheme are stated.According to this algorithm of image retrieval based on salient closed boundary proposed in this thesis, a prototype system is designed and implemented and precision and recall ratio of the retrieval algorithm are tested.The new algorithms are tested by lots of experiments and the results are analyzed. These include:①The experiment results of salient closed boundary extraction illustrate that the new significance measurement function can guarantee the approximation of closed boundary as small as possible and closed area as large as possible. And the results also show that this new algorithm is robust against image noise, rotation, zoom.②The stage experiment results of improved VA-File show that it can greatly enhance the retrieval efficiency with the small influence to the retrieval result.③The experiments of retrieval precision and recall ratio based on the algorithm in this thesis are carried under the image retrieval prototype system. The results show that the algorithm has good performance. In summary, the algorithm proposed in this thesis is effective.