Color in Computer Vision: Fundamentals and Applications

Chapter 4: Pixel-Based Photometric Invariance

Computer vision systems have to deal with widely varying imaging conditions. To obtain robust vision systems, an important property is photometric invariance or the so-called color invariance. Color invariance is derived from color spaces that are more or less insensitive to disturbing imaging conditions such as variations in the light source (both intensity and color), camera viewpoint, and object position.

In the previous chapter, it has been shown that from the RGB color space, several linear and nonlinear transformations can be applied to obtain new color spaces. ¹In this chapter, an overview is given of the color invariant properties of these transformed color spaces. From the dichromatic reflection model, Equation 3.6, it can be derived that the recorded color value at each (pixel) location is highly dependent on the light source characteristics and the object geometry (e.g., the absence/presence of shadows or highlights partly depends on the position of the object with respect to the light source), see Figure 4.1. Many computer vision tasks such as image segmentation and object recognition require stable and repeatable image properties rather than color measurements that are sensitive to imaging conditions. For this purpose, color invariance is needed.

Figure 4.1 Pixel values are highly dependent on the light source characteristics and the object geometry. Assuming Lambertian reflection and white illumination, RGB colors of ...