July 2001

BACK

The first part of his presentation will describe the nature of industrial research in computer vision as well as present an overview of current projects. These projects include: video analysis such as stabilization, tracking and 3-D scene reconstruction; accurate measurement of surfaces from range data using CAD models to guide robust estimation; change detection in aerial imagery using 3-D model context; the integration of natural language and vision using appearance and motion primitives; perceptual grouping based on intensity regions.

The second part of the talk will focus a current project where Dr. Mundy is the principal researcher: computer-aided detection of lung cancer in X-ray CT images. The approach is based on a hierarchy of models ranging from the signal level to models for anatomical structure and cancer lesions. The selection of models based on image evidence and prior knowledge is carried out in a Bayesian framework. Even though the project is at its early stages, the system is capable of distinguishing simple cancer nodules from vascular structure.

Speaker:

Since 1963 Joe Mundy has been a technical staff member of the General Electric Research and Development Center. In 1969 he received his PhD in Electrical Engineering from Rensellaer Polytechnic Institute. His early projects at the Center include: high power microwave tube design, a super-conductive computer memory system, the design of high density associative memory arrays, the application of transform coding to image data compression and the development of an automatic inspection system for lamp filaments. From 1977 until 1982 he was Manager, Visual Information Processing Program. In this capacity he led a number of extensive efforts in the application of machine vision techniques to automatic visual inspection for quality control. In the 1980's Dr. Mundy formed a group to apply Image Understanding algorithms to aerial reconnaissance. He developed a system for aircraft recognition based on a sparse feature, called the vertex-pair, which achieved 98% recognition accuracy in a test on realistic airfield scenes. In the early 1990's, he was co-developer of a new approach to object representation and recognition based on geometric invariance. Over the last five years, he has participated in the RADIUS project, providing algorithms to the RADIUS Testbed System (RTS) which make use of the context provided by a 3D site model. These algorithms include change detection based on various levels of image segmentation and specific object structure matching.

Dr. Mundy's professional activities involve active participation in the areas of pattern recognition and robotics. He was co-chairman of the workshop on industrial applications of machine vision which resulted in a special issue of the IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI) in 1982. He is currently on the editorial board of the International Journal on Computer Vision. He also serves on the NSF advisory board for Artificial Intelligence and Robotics (IRIS). In 1988 he was named a Coolidge Fellow, the highest technical award in GE. He completed a year sabbatical as a visiting fellow at Oxford University during 1988-89. In 1993 Dr. Mundy was co-recipient of the Marr Prize, the leading prize in the field of Computer Vision. He is chairman of DARPA's Image Understanding Environment (IUE) Committee, which has specified and is supervising the development of the IUE. For the last 3 years he has chaired the RADIUS Image Understanding Advisory Committee which has prioritized the potential IU technology sources for RADIUS.

BACK