A Model for Analyzing Dependencies Between Two ICA Features in Natural Images

摘要

In this paper we examine how the activation of one independent component analysis (ICA) feature changes first and second order statistics of other independent components in image patches. Essential for observing these dependencies is normalizing patch statistics, and selecting patches according to activation. We then estimate a model predicting the conditional statistics of a component using the properties of the corresponding feature as well as those of the conditioning feature.