Principal component analysis of sample response to RGB light

In this article we present a principal component analysis (PCA) for data generated by a web camera recording of a two dimensional array of samples illuminated with CRT and TFT screens. In order to create a robust cross-platform identification system useful in bio-analysis we investigate a “buoyant” approach to PCA. Using such approach we assess the relevance of keeping local reference samples to compensate for the spurious angle dependence introduced by TFT screens. We show how local references allow for the detection of such screen dependent angular anisotropy and the construction of a new coordinate space where this effect is considerably reduced. This new coordinate space is built using only the RGB coordinates provided by the camera without resorting to the screen RGB coordinates. Finally we investigate the sensitivity of sample tagging with respect to the number of illuminating colours. We show that 3 quasi-pure R, G and B colours are enough to obtain good separation of data and robust CRT/TFT cross platform tagging.