ABSTRACT
The fundamental problem of vision is the extraction of significant information from images under a tremendously wide range of imaging conditions and consequent variations in image quality. Early efforts to understand the image processing effected in the retinas of biological vision systems consisted of simple linear models of spatial summation and lateral inhibition. More sophisticated filters combine smoothing and sharpening effects and have impulse responses which closely model observed neurophysiological responses. In the early 1970s, the image processing community began to explore one aspect of the problem: the intrinsically multiplicative nature of images as a product of illumination and reflectance. The luminance-dependent behavior of the nonlinear lateral inhibition model is achieved implicitly by the nonlinearity rather than with any explicitly adaptive mechanism. Both homomorphic filters and multiplicative lateral inhibition achieve a relative edge enhancement which is largely independent of local contrast.
