Image Feature Extraction Using Independent Component Analysis

In Independent Component Analysis, one tries to model the underlying data so that in the linear expansion of the data vectors the coeecients are as independent as possible. This often leads to natural features characterizing well the data. In this paper , we present some results on applying Independent Component Analysis to image data. This has become possible by using a recently developed, computation-ally highly eecient xed-point learning rule. The resulting feature masks are sensitive either to lines and edges of varying thickness or to local spatial features and frequencies.