The Center for Bio-Image Informatics faculty have been awarded a new 5 year NSF project that would explore new, fundamental problems in uncertainty analysis while working with image data. The principal investigators on the project include Professors Hollerer, Manjunath, Rose and Singh on the Engineering side and Professors Feinstein, Fisher and Wilson from the biology division.
Images are ubiquitous in scientific applications, and are often the primary source of data for analysis, discovery and hypothesis validation. Typically, these images are processed to extract useful features, which are then used for detecting patterns of interest, building scientific models for data visualization, as well as for new pattern discovery. However, in each of these stages of data analysis there is inherent uncertainty. The premise of this research is that uncertainty is a universal reality in image informatics. Examples may range from remote-sensed imagery of wildfire evolution in ragged terrains, to microscope imagery of suspected breast cancer cells. Powerful models exist for analyzing imagery obtained in either application for determining the current condition (fire front location and propagation rate, cancer cell shape, size and density, etc.) and predicting future evolution. However, errors in parameter estimates are inevitable and inherent to the limitations of instruments and analysis methodologies. The uncertainties in these measures need to be explicitly taken into account in the modeling and recognition stages to optimize threat response strategies. Similar examples exist in virtually all disciplines that use images or video as main data sources. Thus, a highly critical requirement for a broad spectrum of applications is the ability to handle uncertainties in the context of new models and structures for data storage and to consider them in novel data query environments, for search, retrieval, and data mining over such data structures. This is coupled with the need to develop a new generation of visualization techniques to account for implicit uncertainties on multiple levels of analysis. This research project brings together researchers in image analysis, pattern recognition, databases, visualization, and neurosciences in addressing information processing challenges in the specific "testbed" context of bioimaging. The methods and solutions that will be developed cut across disciplinary boundaries and will benefit a wide range of applications. In addition to the basic research, there will be outreach activities that include high-school and undergraduate summer research and workshops that will be hosted at UCSB on this important topic and its implications.
|< Prev||Next >|