Separating corneal reflections for illumination estimation
Neurocomputing
| Hi-index | 0.00 |
When we view a scene through transparent glass, the image is a linear superposition of two images, a real image observed through a glass and a virtual image reflected on it. We can separate the reflections by a polarization and Independent Component Analysis (ICA). Since the image observed through digital camera is non-linearly transformed by gamma correction etc, it may cause error in image processing for image analysis and measurement. The kernel-based methods are effective for such non-linearity. In this paper, we remove the reflections by using Kernel Independent Component Analysis (KICA) and show that KICA is more effective than ICA even if the observed image is non-linearly transformed by camera.