Dynamic Image Analysis Based on Multidimensional Directed Information

Multidimensional directed information (MDI) analysis is a signal processing method to quantify and visualize the causality of multichannel time series as information flows. Although MDI analysis is defined as conditional mutual information among multichannel time series, we have improved the theory of MDI so that it can be applied to other information processing problem. In this study, we propose a new definition of MDI, which is the theory for dynamic image analysis. For example, we expect the new theory helps us to analyze ultrasound image (B-mode movie image) processing, the movement of clouds/gas, the flow of liquids, and so on. We show the formulae of the new MDI definition, which consists of calculations using many subsets of parts of image frames.