This Small Business Innovation Research (SBIR) Phase I project will extend an existing manual patent editing system--adding automatic patent annotation components. The components will include a page segmentation/recognition system adapted to patent figures and an annotation system that links recognized text in the image to corresponding terms in a text document. The immediate technical challenge is providing a high-confidence recognition system despite the sparse nature of the callouts, scan noise, and background line art distractions. This research proposes using the patent specification text to guide the recognition, suggesting what callouts are expected in each figure. The longer-term technical challenge is providing this processor-intensive feature in a way that preserves the responsiveness and extensibility of the browser-based collaborative editor on which it's based. This research proposes to implement the recognition and annotation process as a set of server-side automated participants' which act as remote co-collaborators.
The U.S. economy relies heavily and increasingly upon intellectual property. As patents become more significant in the operations and outcome of U.S. businesses, it becomes increasingly important to assure that the system produces quality patents cost-effectively and expeditiously. The proposed research allows preparers to more easily prepare applications. Patent examiners could use it to automatically detect errors and omissions in applications, improving quality and reducing pendency. Patent researchers could use it to more incisively comprehend, value, and dispute patent prior art. The prototype will, respectively, be evaluated by several attorneys through an application drafting cycle, USPTO examiners via a stand-alone installation, and researchers at a popular search site.
