Study On Blue And White Porcelain Images Retrieval Based On Text And Content

As a typical representative of Chinese porcelain, the blue and white porcelain is a very important kind of cultural relics in China. So it is very necessary and significant to transform tangible materials into digital information and to carry out digital management of blue and white porcelain. How to deal with these digital cultural relic resources and how to take use of these resources have gradually become the focus of study. Image retrieval is a hot spot among all the problems. This dissertation focused on the retrieval of a certain kind of images----the blue and white porcelain images. The study aims at using methods of text retrieval and content based image retrieval to realize image retrieval of these images. The work of this dissertation is as follows:1、This dissertation studied the most classic and widely used ranking algorithm BM25 in text retrieval models and put forward an improved BM25 weighting algorithm according to the characteristics of the blue and white porcelain images. The author classified the vocabulary used in describing blue and white porcelain and gave different category different weights to improve retrieval accuracy. Experimental results showed that the improved BM25 weighting algorithm had a higher accuracy than the original one.2、This dissertation studied the key technologies of content-based image retrieval--feature extraction and similarity measurement. Detailed descriptions of extraction algorithms including color histogram, gray level co-occurrence matrix, Tamura texture feature and Gabor texture feature were explained. Then these four kinds of image features were used for image retrieval respectively. This dissertation put forward a multi-feature fusion method for blue and white porcelain image retrieval by combining color feature with texture feature. Experimental results showed that multi-feature retrieval, which had a higher accuracy, performed better than single feature retrieval.3、Text retrieval and content based image retrieval were combined to realize blue and white porcelain image retrieval. Users used text retrieval in the first place, then the results were re-ranked according to the feature similarity between images, which improved the retrieval performance.