Finding Outliers in Image Data and Video

Surface, image and video data can be considered as functional data with a bivariate domain. It is well known that classical statistical techniques to detect outlying surfaces or to flag outlying parts of a surface are not trustworthy as the analysis itself may be distorted by the outliers. We propose a new display based on the mean and the variability of the degree of outlyingness at each grid point. A rule is constructed to flag the outliers in the resulting functional outlier map. Heatmaps of their outlyingness indicate the regions which are most deviating from the regular surfaces. Using projection pursuit techniques, the method is applicable to univariate as well as multivariate surface data. To illustrate the performance of the method, it is applied to fluorescence excitation- emission spectra after fitting a PARAFAC model, to MRI image data which are augmented with their gradients, and to video surveillance data.status: publishe