Synthesizing a Color Algorithm from Examples

TLDR

The algorithm resembles Land’s recent lightness model and behaves like filtering through center‑surround receptive fields in chromatic channels, a property shared by a class of early‑vision algorithms that can be synthesized from examples. The study aims to automatically synthesize a lightness algorithm that separates surface reflectance from illumination using example pairs of intensity signals and target reflectance images. The synthesis uses optimal linear estimation, assuming the input‑to‑output operator is linear, which is sufficient for this class of algorithms. Back‑propagation and other learning methods fail to produce a noticeably better lightness algorithm than the linear estimation approach.

Abstract

A lightness algorithm that separates surface reflectance from illumination in a Mondrian world is synthesized automatically from a set of examples, which consist of pairs of input (intensity signal) and desired output (surface reflectance) images. The algorithm, which resembles a new lightness algorithm recently proposed by Land, is approximately equivalent to filtering the image through a center-surround receptive field in individual chromatic channels. The synthesizing technique, optimal linear estimation, requires only one assumption, that the operator that transforms input into output is linear. This assumption is true for a certain class of early vision algorithms that may therefore be synthesized in a similar way from examples. Other methods of synthesizing algorithms from examples, or "learning," such as back-propagation, do not yield a significantly better lightness algorithm.