We present a spatio-temporal filter for real-time noise reduction of strongly corrupted X-ray image sequences. It possesses efficient noise reduction while, at the same time, preventing typical artifacts of state-of-the-art methods. Decisive for these features are, in particular, innovative motion detection as well as noise-adaptive filter parametrization. Motion detection based on twofold signed binarization proved to be a powerful method for pixelwise separation of motion and strong noise. Drawbacks of threshold determination by Euler curve analysis as applied previously were eliminated by integration of signal-dependent noise estimation.
