基于文本的科技论文图像检索

he necessity of building image retrieval system on scientific documents database is analyzed and two related problems are studied . One is that an image extraction method is proposed to extract images from scientific documents . In this method ,the documents are transformed to document images ,and then color histogram ,1-order color moment and 2-order color moment are used to find out content images from document images and a precision value of 94.3% is obtained . The other is that rule-based related text extraction method is proposed . Different combinations of title , abstract , keywords and surround text are used to index the content images .The experiment result shows that using the title and surround text to index images results in best retrieval performance .