THE Video Time Density Function and Its Applications on Video Content Analysis
Xiao-Ping Zhang, Ph.D., P.Eng., M.B.A., SMIEEE
Professor, Director of Communication and Signal Processing Applications Laboratory (CASPAL)
Program Director of Graduate Studies, Department of Electrical & Computer Engineering
Ryerson University, Canada
In this talk, first we show that video temporal dynamics contain important semantic information by an example. We then present a new video temporal density function (VTDF) concept to describe the time density of video activities. We build VTDFs based on the inter-frame mutual information in a specific video feature space. Based on the VTDF, we can formulate many video navigation and summarization problems as a generic nonuniform sampling problems. VTDFs enable us to develop new temporal quantization-based methods for video content analysis. VTDFs built on different video features can capture different types of video semantics. We show some preliminary application examples on VTDF based rapid video navigation and summarization methods and demonstrate their effectiveness.