Career: Non-parametric Multi-scale Statistical Models for Natural Signals and Images

点击翻译按钮获取中文摘要

This CAREER award supports the study of image processing and vision. Vision is perhaps the most complex and powerful capability that we possess. Intriguingly vision develops gradually as we grow from infancy to adulthood. This research focuses on two related questions: What is the fundamental structure of visual images that allow this type of learning? What algorithms can be used to learn this structure? Answers to these questions have bearing on the recognition of complex visual patterns, the generation of computer graphics images, and the study of human perception. The technical emphasis of this effort is based on the observation that the distribution of natural images is far from uniform. On the contrary, real images have complex and important statistical structure that can be exploited for image processing, recognition and analysis. Based on these insights, the investigator has developed a non-parametric multi-scale statistical model for images that can be used for recognition, image processing and in a ``generative mode'' to synthesize high quality textures. In this effort the theoretical underpinnings of these algorithms as well as the application to other processing problems are studied. In addition, the award is supporting the development of a class and textbook on the statistical structure of natural images. While emphasizing fundamentals, the course draws together advances in computer vision, image processing, digital television, image compression, the psychology of vision, and the neural basis of vision.