Four high-frequency components are extracted from video signals: the first from a blanking interval; the second from an image-carrying period between blanking intervals; and the third and fourth from the video and one-line and -frame delayed signals, respectively. Statistical processing is performed to obtain absolute values of the second to fourth component levels per pixel and the number of pixels of the components per image per absolute level. A noise detecting signal is generated based on the first component level irrespective of the processing when the level is higher than a predetermined level, if not, first to third levels are obtained for the first to third components, respectively, each having the smallest number of pixels among levels other than zero each having a larger number of pixels than zero for the components. The detecting signal is generated based upon the lowest or second lowest level among the first to third levels.