The influence of the number of signals on the clustering aspects of independent component analysis

Abstract

The human visual system uses a statistical approach calles Independent Component Analysis (ICA) to detect objects in the view field. Based on the idea that, give the huge number of objects, this ICA normally should be incomplete, we analyze which influence has the number of sources on the results of the incomplete ICA. We show that an incomplete ICA clusters the underlaying sources based on their similarity of their columns in the mixing matrix and though, applied after filtering, can be used to detect objects in natural scenes.