This research project will develop methodologies in multivariate statistical analysis for the purposes of applications to statistical image learning, incomplete data, regularized discriminant analysis, and related areas. The investigators will develop exact inference for multiple population comparisons, principal components, imputation procedures, clustering and mixture modeling techniques for large-scale image classification, and regularization methods for statistical image models. The results of the project will have impact on image retrieval and annotation, and statistical methods for analyses of data arising from national panel surveys, social and behavioral research, toxicology research, and any area in which multivariate incomplete data commonly appear. Computer software packages for algorithms and analytical tools developed in the project will be documented for public access. Owing to cutting-edge trends in wildlife research, environmental sciences, social and behavioral research, image analysis, and related areas, new statistical and probabilistic ideas are needed to analyze statistical data at small sample sizes. The investigators will develop statistical formulas and computational algorithms for analyzing such data at small sample sizes. Image searches provided by major search engines, such as Google and Yahoo!, rely on textual descriptions of images found on Web pages containing the images and file names of those images. This project will advance the technology of computer-assisted annotation of images by words and hence enhance the visibility of images on the Internet. Other benefits of the project to the society include outreach activities in which the investigators will be involved in summer school programs for high-school students and the development of publicly-available image analysis and discriminant analysis software.
|Effective start/end date||7/1/07 → 12/31/10|
- National Science Foundation: $290,000.00