Subpixel anomalous change detection in remote sensing imagery

A machine-learning framework for anomalous change detection is extended to the situation in which the anomalous change is smaller than a pixel. Although the existing framework can. be applied to (and does have power against) the subpixel case, it is possible to optimize that framework for the subpixel case when the size of the anomalous change is known. The limit of inLntesimally small anomaly turns out to be well deLned, and provides a new parameter-free anomalous change detector which is effective over a range of subpixel anomalies, and continues to have reasonable power against the full-pixel case.