| Cartoon industry went by name of “Somkeless industry”,which play a more and more important role in increasing the development of economy around the world. With the development of cartoon industry, on one side all the famous companies in cartoon all accumulate a great deal of cartoon material, on the other side more and more cartoon materials are lay aside and could not use effectively, all the reasons cause the cartoon industry come to a deadlock. Under this background, we carried out study and experiments on cartoon material retrieval pivotal technology, and constructed a cartoon retrieval system based on content, which would reduce cost and improve competition,meanwhile resist the cartoon industry abroad and culture invasion abroad.Through on studying the existing algorithm, we made technical route, combined with cartoon materials’ characteristics. With the study on existing algorithms, combined with its own characteristics of cartoon material on having more saturation and value, we used 96-color non-uniform quantization algorithm based on HSV to extract color features, which subdivided saturation and value meanwhile increased their weights;Second,compared with nature textures, cartoon textures are more rough and changes slowly, so we used gray level co-occurrence matrix’s inverse-gap,uniformity and correlation to extract texture features; third we used Hu matrix to extract shape features.Last, we used support vector machine to construct separator to realize automatic dimensioning and classification of semantic. And we constructed a cartoon material retrieval system. The platforms for retrieval system are Windows Advanced Server 2003 and SQL Server 2003, and the development environment is visual C++6.0.By experiment, we approached a satisfactory result. The accuracy of image retrieval reached to 84%. Through the system, it can help user to realize cartoon material search, which would improve the utilization rate of cartoon material and the production rate,meanwhile reduce the cost. |
PDF Full Text Request