Printer-friendly versionSend to friend

Command, Control and Interoperability Center for Advanced Data Analytics (CCICADA)

Standard Model

CCICADA Model

Raw data of texts and images can be difficult to accurately interpret unless they pass through a “tool.” The human eye and brain represent just such a tool by providing a natural, unambiguous format for interpreting texts and images. In much the same way scientific tools are being developed to filter and improve these interpretative functions of illustrative material.

The challenge of developing interpretive tools is being taken up by the Command, Control, and Interoperability Center for Advanced Data Analysis (CCICADA). Led by Professor David Forsyth, researchers at the University of Illinois at Urbana‐ Champaign are using approximate inference methods, together with structured learning, to produce improved parses of human figures. According to Dr. Forsyth, the research team will use the text found near images together with the images to produce (a) improved interpretations of those pictures and (b) methods to predict interpretations or textual annotations for other images that do not have text. Human parsing is the problem of determining automatically where a person’s arms, legs, body and head are in an image. CCICADA researchers are using support vector machine ranking methods to exploit similarity data so that similar objects can be used to train a system to recognize a particular object.

The breadth and volume of information that is archived and streams across networks needs additional tools for improving filtering and interpretation use in a viable, simple object recognition system. To that end, David Forsyth and his team are developing and revealing methods, which show that human parsing in single images can be significantly improved. Furthermore, this similarity in approach can also produce recognizers that work when there is very little training data.