Research Of Engineering Drawings Retrieval Technology Based On Multiple Feature And Relevance Feedback

With the rapid development of multimedia and internet technology, the application of engineering drawings is more and more extensive in production department of the enterprises. How to retrieve the necessary information from large amount engineering drawings quickly and efficiently needs to be solved urgently. Combined with the characteristics of engineering drawings, applying the image processing technology, the key techniques of engineering drawings retrieval are deeply researched in this thesis.Firstly, the research situations of several typical content-based image retrieval systems and the key techniques of them are analyzed. According to the retrieval requirements of the engineering drawings, an engineering drawings retrieval system based on multiple features and relevance feedback is designed. To make up the limitation of the retrieval method using a single feature, the engineering drawings retrieval method utilizing density feature and shape feature is presented. Invariant moments, improved direction code and two-dimensional Fourier descriptor are separately utilized to describe the shape feature of the engineering drawings from different aspects. Density feature can effectively describe the object distribution of different regions in the engineering drawings.To further improve the retrieval performance of the system, the algorithm of relevance feedback based on SVM is improved and applied in the engineering drawings retrieval effectively. The sample insufficiency problem in the beginning of retrieval process is resolved by accumulating the feedback samples after each iteration. After establishing the optimal classification, the twice distance of engineering drawings is computed to reinforce the semantic discrimination. Therefore, the shortcomings of pattern classify method in the relevance feedback is resolved.Finally, the implementation of engineering drawings retrieval system based on multiple features and relevance feedback is presented. We introduce the application into the engineering drawings retrieval of Product Data System in certain production enterprise initially and achieve a series of good retrieval results.