The problem of tracking articulated 3D human motion in video sequences is a challenging one since it involves extracting relevant information from each frame and then exploiting the information in a principled manner determine the pose of each limb. We study the methods of learning image statistics, how humans look and move images and see how these statistics can be used to track humans. Two di...

In this paper we propose a novel method of scene classi cation, based on the idea of mining emerging patterns between classes of images, represented in a symbolic manner. We use the 9DLT (Direction Lower Triangular) representation of images, which allows to describe scenes with a limited number of symbols, while still capturing spatial relationships between objects visible on the images. We sho...

There exist numerous psychophysical paradigms for which performance varies with location of stimulus presentation within the visual field. The following considers potential bases for visual anisotropies, considering the possibility of a statistical basis for such effects in cases where basic sensory asymmetries are present. In particular, the relationship between scene statistics and both upper...

We propose a computational auditory scene analysis (CASA) model for monaural speech segregation. It deals with low-frequency and high-frequency signals differently. For high-frequency signals, it generates segments based on common amplitude modulation (AM) and groups them according to AM repetition rates. This model performs substantially better than previous CASA systems.