The work aims to maximize the invisibility of watermark while ensuring the algorithm robustness, adaptively select the embedded area of the watermark and embed the watermark signals of different intensities according to the image features. By means of blocked image, the area with the most complicated texture was selected according to the various characteristic parameters of gray-level co-occurrence matrix quadratic statistic for the carrier image. Textural features were defined by fuzzy C-means (FCM) cluster. Then, the texture intensity of all blocks was analyzed by grey level histograms. Optical holography encrypted watermark signals with diverse intensities were adaptively embedded into the diagonal component of image wavelet domain. The invisibility of watermark was significantly improved (PSNR=49), and the watermark information extracted when the image was subject to such operations as filtering, noise, cutting and rotation could still be clearly recognized. The proposed algorithm has greatly improved the invisibility of watermark while ensuring the robustness and security.