Analysis of video images from a gas-liquid transfer experiment: a comparison of PCA and PARAFAC for multivariate image analysis

The use of chemical imaging is a developing area which has potential benefits for chemical systems where spatial distribution is important. Examples include processes in which homogeneity is critical, such as polymerizations, pharmaceutical powder blending and surface catalysis, and dynamic processes such as the study of diffusion rates or the transport of environmental pollutants. Whilst single images can be used to determine chemical distribution patterns at a given point in time, dynamic processes can be studied using a sequence of images measured at regular time intervals, i.e. a movie. Multivariate modeling of image data can help to provide insight into the important chemical factors present. However, many issues of how best to apply these models remain unclear, especially when the data arrays involved have four or five different dimensions (height, width, wavelength, time, experiment number, etc.). In this paper we describe the analysis of video images recorded during an experiment to investigate the uptake Of CO2 across a free air-water interface. The use of PCA and PARAFAC for the analysis of both single images and movies is described and some differences and similarities are highlighted. Some other image transformation techniques, such as chemical mapping and histograms, are found to be useful both for pretreatment of the raw data and for dimensionality reduction of the data arrays prior to further modeling. Copyright (C) 2003 John Wiley Sons, Ltd.