A comparison of registration metrics for correlating non-compatible images

In the field of pattern recognition it is often necessary to locate objects in a given image by referring to their position in another image of the same scene. Thus, in image processing it is important to accurately and efficiently register (match) two images of the same scene even when they may be recorded by separate sensors under different conditions.

In a search for an improved registration measure, in this work various metrics were used to register a set of compatible and non-compatible images. Specifically, the standard metrics, including the direct cross-correlation (DOC), fast Fourier transform (FPT) and mean absolute different (MAD) metrics, were compared to the new thresholded difference (TD) metric. In the case of compatible images, the TD metric produced a sharper matching peak than the other pixel-by-pixel methods investigated. For non-compatible images representing different spectral bands, times of day and scale, the TD metric, with a suitable threshold, gave results which were generally as good as or better than the other metrics used.