Robust Two-Sample m-Interval Partition Detectors with Application to Image Processing

The theory of the m-interval partition detector is proposed for the two-sample test. The locally most powerful scores are derived for the test, and the performance of the partition detector is investigated using these scores. The detector operates at near optimum level for the underlying noise distributions, and maintains its robustness in the changing intrinsic noise environments. In addition, the detector is applied to retrieve edges from noisy images corrupted by Gaussian noise and salt-and-pepper noise.