A blind watermarking method based on binarization computational ghost imaging was proposed. In this method, a watermark image is firstly encoded by the encryption system based on computational ghost imaging, and then the ciphertext is binarized and embedded in the discrete cosine transform domain of a host image to realize watermark embedding. The process of watermark extraction and reconstruction is the inverse process of embedding and encryption. The watermark is obtained by extraction key and decryption key. Simulation results show that this watermarking technique has good imperceptibility. When α=10, the watermark is also imperceptible and the peak signal-to-niose ratio of the host image containing watermark is above 38 dB. In addition, this method also has some fault tolerance ability. When the error rate of the extracted encrypted data reaches 20%, the reconstructed watermark information can still be distinguished and recognized. Compared with the traditional computational ghost imaging, binarization for the ciphertext data brings convenience for watermark embedding, but does not bring serious deterioration to reconstructed image, and the difference of correlation coefficient is less than 0.1. In the process of watermark extraction, it is unnecessary for the original host image, so it belongs to a blind extraction method.