Color in Computer Vision: Fundamentals and Applications

Chapter 7: Photometric Invariance by Machine Learning

With contributions by José M. Àlvarez and Antonio M. López

As shown in the previous chapters, the choice of a color model is of great importance for many computer vision algorithms, as the chosen color model induces the equivalence classes to the actual algorithms. ¹As there are many color models available, the inherent difficulty is how to automatically select a single color model or, alternatively, a weighted subset of color models producing the best result for a particular task. The subsequent hurdle is how to obtain a proper fusion scheme for the algorithms so that the results are combined in a proper setting. In the previous chapters, physical reflection models (e.g., Lambertian or dichromatic reflectance) are used to derive color invariant models. However, this approach may be too restricted to model real-world scenes in which different reflectance mechanisms can hold simultaneously. Instead of modeling the world by a single reflection model, we now focus on how color invariance can be obtained by machine learning.

The learning process is based on the selection of positive examples (e.g., colored image patches of a certain object to be recognized) to obtain color invariant ensembles. Of course, the training examples should include a broad range of pixel values capturing all possible imaging conditions under which the object can be captured. Using these training samples, the aim is to arrive at color ensembles that yield ...