Unsupervised texture segmentation by Hebbian learnt cortical cells

Abstract

In this letter, principal component analysis (PCA) type Hebbian learning is proposed as a mechanism by which orientation and frequency selective channels can be tuned to extract maximal information from within an image. Using these channels, unsupervised texture segmentation is performed using texture edge detection. Preliminary results are presented for a variety of synthetic, perceptual and naturally occurring textures. Finally, possible applications are suggested for the method together with areas of future extension of the method.