Abstract

We applied redundancy reduction and image compression methods to image signals in space and color to derive a filtering operation that simultaneously reduces correlation in both these dimensions. The hybrid filtering operation is based on linear predictive coding (LPC) in space in conjunction with an eigenvector (Karhunen-Loeve) transformation in color. The resulting filter has a distinct major component associated with a luminance response, while further components are associated with chromatic responses. The luminance component has a high spatial frequency bandpass form with a large dynamic range and the chromatic components are spatially low pass and contain chromatic differences with a small dynamic range. The filter transforms the image into separate components that contain primarily spatial information and primarily color information. These filter components correlate with a high spatial frequency cutoff luminance mechanism and low spatial frequency cutoff chromatic mechanisms in the visual system. These spatiochromatic filtering mechanisms can be implemented by combining simple color opponent receptive fields. This suggests that the receptive field architecture is used as a means of redundancy reduction and image compression by generating transformations that separate an image into components that have (i) high spatial sensitivity and low chromatic content and (ii) high chromatic sensitivity and low spatial acuity.

© 1989 Optical Society of America