Analysis of data that are derived from advanced image acquisition techniques, gains more influence in the image processing industry. One major issue in the visual inspection in the industry is the detection of anomalies. Rather than detecting anomalies by describing them by features or detecting them by explicitly describing the non-anomaly case, the auto-regressive models provide a way to eliminate expected pattern and emphasize the not expected - the anomalies.
This paper introduces a new class of auto-regressive (AR) models that can handle data which contain different kinds of modalities, where the modalities represent different aspects of the inspected surface. The theoretical background of the AR models are presented, explained and analyzed.