Abstract
Bandpass filtering of an image is commonly used as the first step toward edge detection. However the bandpass signal retains additional information (edge primitives) from which it is possible to recover a more nearly complete representation of the original scene. This method can lead to a richer scene description in machine-vision applications and to new coding schemes for applications that require high data compression. We consider two different bandpass filters, both of which are based on models of retinal processing in human vision. One is the spatially invariant Laplacian of Gaussian (∇2G), and the other is the spatially variant intensity-dependent spatial summation (IDS). Both filters preserve significant information beyond the location of edges. The ∇2G filter preserves information from which we can extract the change of intensity across the edge boundary. The IDS filter preserves information from which we can extract both the actual intensities and, often more important, the reflectance ratio across the edge boundary. The recovery of the relative reflectance is highly robust to temporal and spatial changes in illumination (e.g., shadows). In this paper we present the method of recovering the intensity and reflectance representations from the bandpass-filtered signal and assess the accuracy, stability, and resolution of these representations.
© 1990 Optical Society of America
Full Article | PDF ArticleMore Like This
Shahriar Najand, Douglas Blough, and Glenn Healey
J. Opt. Soc. Am. A 13(7) 1305-1314 (1996)
John G. Daugman
J. Opt. Soc. Am. A 5(7) 1142-1148 (1988)
Timothy J. Schulz and Donald L. Snyder
J. Opt. Soc. Am. A 9(8) 1266-1272 (1992)