The Image Formation and Processing group is concerned with research issues related to the acquisition, manipulation, and synthesis of images. The many research topics fall into three broad categories: computerized imaging; image-video transmission, storage, and manipulation; and image and scene modeling and analysis.
Group Leader: Thomas S. Huang
Administrative Support Staff: Sharon Collins
Computerized Imaging: This research deals with the signal-processing techniques and algorithms required to form 2-D or 3-D images from multiple-sensed data sets. Examples include X-ray tomography, magnetic resonance imaging, and synthetic aperture radar. The objective is to obtain high-quality images from noisy or incomplete data. Applications include improved image formation in medical and industrial CT and MRI scan instruments, higher-quality images of the terrain from airborne or spaceborne radar, and forming airport runway images when the visibility is poor.
Image-Video Transmission, Storage, and Manipulation: The goal of this research is to find representations of images and video that require only a small amount of bits to transmit or store, yet are easy to find in a database. The group studies image and video compression by using approaches ranging from wavelets and fractals to ideas from pattern recognition and computer vision. Also under study are methodologies and techniques for image-video indexing and editing. Research in visual perception is essential to this effort because, in many applications, humans view the images after processing, so the subjective quality is an important performance criterion. Applications include video phone, teleconferencing, and multimedia databases. Increasingly, this research has combined image/vision with audio/speech; e.g., in the video indexing project, the group is using both visual and audio cues to derive semantic labels for video shots.
Image and Scene Modeling and Analysis: A major research effort is the modeling, analysis, and visualization of human movement. First, good 3-D models of the human head/face, hand, and body are constructed from multiple-sensed 2-D images, and then the models are used to do analysis and synthesis. Applications include vision-based human-computer interfaces, very low bitrate model-based video compression for videophone and teleconferencing, and computer recognition of American Sign Language.
Vision- and image-based techniques will play a key role in most human-computer intelligent interaction scenarios, such as in collaborative manufacturing product prototyping in a virtual environment, thus making the interaction between people and virtual environments more natural and efficient.