Through analysis of dither modulation algorithm of signal processing and the masking effect of human visual system, a DCT domain blind watermark model was proposed. Watermark preprocessing, embedding, extraction, and F norm-based detection were realized for binary image watermarks in MATLAB simulation environment. The imperceptibility of watermark was improved through automatically adjusting the quantization step during watermarking. Blind detection of watermark was realized without the original host image and algorithm complexity was reduced by leaving out inverse discrete cosine transform process. The simulation results showed that the scheme has stronger robustness against conventional attacks such as higher JPEG compression, cropping, noise overlap, multiple watermarking, etc. This scheme can be applied in digital copyright protection and secret communications.