In order to balance the algorithm’s robustness and the watermark’s transparency better, a blind digital watermarking algorithm based on SVD decomposition was proposed. First, divided the original image into a number of 8*8 size blocks; then execute DCT transform on each block; next, execute SVD decomposition on each 2*2 size block of DC coefficients set with the purpose of get-ting the corresponding singular value matrix; finally, complete the watermark embedding by quali-fying the largest singular value using a step value Q based on the principle of parity quantification. The algorithm extracts DC coefficients as carriers of watermark image from the DCT coefficients domain, ensuring the robustness of this algorithm. Meanwhile, the transparency of the watermark image is increased by using the result of SVD transforms performs intrinsic properties of images ra-ther than the visual characteristics of images. Further more, this paper theoretically analyzes the in-fluence of the quantization step Q on the image quality, and selects the optimal Q value according to the result of the experiments. Experiments show that the algorithm can effectively resist JPEG compression, median filtering, noise and other conventional tampering operations, watermark can be extracted without the participation of the original image, achieving the blind detection of water-mark.