A new approach to clustering and object detection with independent component analysis

Abstract

It has previously been suggested that the visual cortex performs a data analysis similar to independent component analysis (ICA). Following this idea we show that an incomplete ICA, applied after filtering, can be used to detect objects in natural scenes. Based on this we show that an incomplete ICA can be used to efficiently cluster independent components. We further apply this algorithm to toy data and a real-world fMRI data example and show that this approach to clustering offers a wide variety of applications. © Springer-Verlag Berlin Heidelberg 2005.