Difference Detection, Tradeoffs Between Detection Probability and Processing Complexity

The detection of a difference between two n-dimensional vectors based on a noisy measurement of each vector is important for surveillance, quality control, motion detection, and other applications. Because of cost or complexity, an arbitrary reduction from n to p dimensions may be required before measurements are performed. Using a Gaussian noise assumption, the tradeoffs between detection probability and p, which we take as the measure of processing complexity, are investigated.