Abstract

It is well known that the receptive fields of simple cortical cells can be modeled by Gabor functions. This can be a new and promising way of facing the texture problem. In our presentation we show how the responses of the neurons in the visual cortex can be used as texture descriptors in image analysis tasks—mainly segmentation and classification. We have used a simple Gabor scheme with a pyramid implementation in the spatial domain. Only four orientation channels and four frequency channels have been used. We have also searched for descriptors that are invariant under rotation and scale changing. Equivalent texture analysis results may be expected with other similar schemes, such as the cortex transform. Because this Gabor coding is not complete, there will be some errors during the reconstruction. These errors have been studied, including the effect of missing frequency channels. In spite of these errors, the resulting reconstructed images have good visual quality, except for some typical visual illusions. Because these illusions also appear in the visual system, a similar scheme might be used as a model of the visual system.

© 1990 Optical Society of America