Color and Active Infrared Vision: Estimate Infrared Vision of Printed Color Using Bayesian Classifier and K-Nearest Neighbor Regression

Abstract

Speaking of active infrared vision, its inability to see physical colors has long been considered as one major drawback or something everybody has paid no attention to until very recently. Looking at this color blindness from other perspective, we propose an idea of a novel medium whose visibilities in both visible and active infrared light spectrums can be controlled, enabling vision-based techniques to transform everyday printed media into smart, eco-friendly and sustainable monitor-like interactive displays.

To begin with, this paper observes the most important key success procedure regarding the idea—estimating how physical colors should look like when being seen by an active infrared camera. Two alternative methods are proposed and evaluated here. The first one uses Bayesian classifier to find some color-attribute combinations that can precisely classify our sample data. The second alternative relies on simple weighted average and k-nearest neighbor regression in two color models—RGB and CIE L*a*b*. Suggesting by experimental results, the second method is more practical and consistent at different distances. Besides, it shows likelihoods of the model created in this work being able to estimate infrared vision of colors printed on different material.