Analysis of clipping effect in color images captured by CCD cameras

Abstract

CCD cameras have a limited dynamic range that constrains the brightness of the incident light that can be quantified. If the ray of incident light is too intense, the sensor saturates. This color clipping effect is a common problem in computer vision and it can become specially difficult when dealing with specular objects against a low-intensity background. In this paper, we present a method for analyzing such clipping effects appearing in color images. Using an averaging technique to estimate the color of the illuminant, we define two types of axes in the RGB color cube: the illumination axis and the clipping axis. Our study concludes the following: 1) the clipped pixels from a dielectric object form one or two lines, depending on the number of color channels on which the clipping effect takes place; and 2) these lines are parallel to the clipping axes. These two properties allow for a framework for a color-based segmentation that works even in the presence of saturated (specular) regions in the image. Moreover, the captured images can now be obtained under a wider variation of illumination conditions.