From Information Scaling of Natural Images to Regimes of Statistical Models

• Activity Number: 435
• Type: Invited
• Date/Time: Wednesday, August 10, 2005 : 2:00 PM to 3:50 PM
• Sponsor: IMS
• Author(s): Yingnian Wu*+ and Song-Chun Zhu and Cheng-En Guo
• Companies: University of California, Los Angeles and University of California, Los Angeles and University of California, Los Angeles
• Keywords: Natrual Images ; Information scaling ; Markov random fields ; Wavelets ; Sparse coding ; Primal sketch

Abstract

Computer vision can be considered a highly specialized data collection and analysis problem. We need to understand the special properties of image data to construct statistical models for representing various image patterns. In this paper, we study one fundamental perspective of image data of natural scenes---visual objects and patterns can appear at a wide range of distances or scales. The same visual pattern appearing at different distances or scales produces different image data with different statistical properties. In particular, we show the entropy rate of the image data of the same visual pattern changes over the the viewing distance (as well as the camera resolution). Moreover, the inferential uncertainty of the underlying visual pattern changes with viewing distance, too. We call these changes information scaling. From this perspective, we examine both empirically and theoretically two prominent and yet largely isolated research themes in image modeling literature: wavelet sparse coding and Markov random fields.