The constantly increasing number of yearly patent applications leads to an increased workload at the patent offices, raising the need for applications and systems that help the patent expert to faster evaluate an application. Up to now, all the software solutions for mining of patents use textual cues only. Hence, important information contained in patent drawings is inaccessible to automated analysis and retrieval. As a result drawings are sometimes misused to "hide" information and to trick full text search systems. Thus the complex task of manually analysing and comparing drawings is left to the expert, which in turn leads to high financial costs. This situation demands the development of innovative algorithms for the mining of patent images (e.g. information extraction, linking and image based search). This will help patent offices to cope with the ever increasing number of filings.
To make the information buried in patent images accessible, the project developed novel methods for:
- automatic interlinking between patent text and drawing parts by sub-part segmentation and label identification,
- efficient search for similar sub-parts of a drawing within a large scale patent database,
- an interactive user-guided search with manual feedback provided by the patent expert.
The developed techniques aim for application to a variety of technical drawings, such as flow charts, block diagrams, time charts and graph plots. Furthermore, the project evaluated the proposed methods against a real life corpus of patents and patent images. They are integrated in a prototype application to show their possibilities and benefits.
This project was coordinated by JOANNEUM RESEARCH. JRS and TUW were research partners, m2n acted as industrial partner.