As high-resolution digital cameras become more affordable and widespread, high-quality digital images become ever more available and useful. There is an urgent need to support more effective image retrieval over large-scale image archives. In this Small Grants for Experimental Research (SGER) project, a novel framework for image content representation will be developed based on using salient objects to characterize middle-level image semantics. For salient object detection, automatic image segmentation will be integrated with image region classification. Also, hierarchical image classification will be incorporated to enable concept-oriented image database indexing and retrieval through exploring the underlying relationships among different semantic image concepts. Hierarchical finite mixture models will be presented for multi-level semantic image concept modeling, image classification and database indexing, and an adaptive Expectation Maximization (EM) algorithm will be developed for model selection and parameter estimation. Finally, a novel framework for statistical image modeling will be developed to enable partial image matching in the procedures for classification and retrieval. The research will lead to useful tools that could significantly impact computer vision, image database management, digital library or museum management, and image data mining for a wide range of applications. The project will provide a good environment for interdisciplinary education in information technology that bridges the gap between traditional areas of computer vision, databases, visual information retrieval, and machine learning to benefit students of all levels.
Project web page: www.cs.uncc.edu/~jfan/project3.html