Automated analysis of images in documents for intelligent document search

We use images to present a wide variety of important information in documents. For example, two-dimensional (2-D) plots display important data in scientific publications. Often, end-users seek to extract this data and convert it into a machine-processible form so that the data can be analyzed automatically or compared with other existing data. Existing document data extraction tools are semi-automatic and require users to provide metadata and interactively extract the data.

 Image classification

Automatic image classification is often an important step in content-based image retrieval and annotation. Prior efforts model the retrieval and annotation problems as automatic classification of images into classes corresponding to semantic concepts. Visual features and modeling techniques have attracted significant attention. Textural features, color features, edge features, or combinations of these features have been developed for classifying images. Chapelle et al, used support vector  machines to improve the histogram-based classification of images. Li et al. utilized context information of image blocks, i.e., statistics about neighboring blocks, and modeled images using two-dimensional hidden Markov models to classify images.Maree et al. proposed a generic image classification approach by extracting subwindows randomly and using supervised learning. Yang et al.  designed a method to learn the correspondence between image regions and keywords through Multiple-Instance Learning (MIL).

Image analysis

Recognition and interpretation of graphics, such as engineering drawings, maps, schematic diagrams, and organization charts, are important steps for processing mostly graphics document images. Yu et al. developed an engineering drawing understanding system for processing a variety of drawings. The system combines domain-independent algorithms, including segmentation and symbol classification algorithms, and domain-specific knowledge, for example a symbol library, in the processing of graphics. Okazaki et al. proposed a loop-structure-based two-phase symbol recognition
method for reading logic circuit diagrams. Blostein et al. summarized various approaches to diagram recognition. Futrelle et aldeveloped a system to extract and classify vector format diagrams in PDF documents. Shao et al. designed a method for recognition and classification of figures in vector-based PDF documents.