Data Analysis on Multivariate Image Set

Abstract

An image set can include not just the images themselves but also the extracted features, metadata and so on. For example, x-ray images obtained from synchrotron beamlines are large-scale highdynamic-range data depicting a variety of material properties after incorporating scientific analysis results. Previously, we presented a framework MultiSciView as an image set visualization and exploration system for x-ray scattering data. This tool is general enough to deal with any multivariate images. In this work, we aim to complement it with a set of data analysis modules. First, we present feature analysis by proposing a new correlation metric to reduce the data redundancy. Then we encode each image as a high dimensional vector and analyze the patterns hidden in the image set. Finally, we add an auxiliary visualization to plot the average and entropy images of the interested subset. We conducted one case study to show that our system can effectively analyze the image set, identify preferred image patterns, anomalous images and erroneous experimental settings. Eventually a better comprehension of the material nanostructure properties can be achieved.