Use of inferential statistics to estimate error probability of video watermarks

Errors in video watermark detection can cause serious problems, such as erroneous indication of illegal copying and erroneous copy control. These errors could not, however, be eliminated because watermarked pictures are subjected to wide varieties of image processing such as compression, resizing, filtering, or D/A or A/D conversion. Estimating errors of video watermarks is therefore an essential requirement for electric equipment that is to use copyright and copy-control information properly. This paper proposes a video watermarking method that estimates error probability from each watermarked frame at hand after image processing by using the expectation-maximization algorithm from inferential statistics. The paper also proposes a reliable detection system of video watermarks by using the proposed method. Experimental evaluations have shown that the new method can be used reliably with the margin factor and can be widely used in electric equipment as well as content-distribution systems.