Automatic Alignment Detection and Correction in Infrared and Visual Image Pairs

Previously well aligned image sensors, mounted on the same camera, might become misaligned depending on external vibrations. It is of interest to be able to automatically detect and correct for this misalignment, and to separate the deviation into pointing- and/or parallax errors. Two methods were evaluated for this purpose, an area based image registration method and a feature based image registration method. In the area based method normalized cross-correlation was used to estimate translation parameters. In the feature based method, SIFT or LIOP descriptors were used to extract features that were matched between the two image modalities to estimate transformation parameters. In both methods only image points that were in focus were extracted to avoid detection of false alignment deviations. The results indicate that the area based image registration method has potential to automatically detect and correct for an alignment deviation. Moreover, the area based method showed potential to separate the deviation into pointing errors and parallax errors. The feature based method was limited to specific scenes but could be used as a complement to the area based method in order to additionally correct for rotation and/or scaling