In order to protect the copyright of color digital images in the current Internet environment, this paper proposes a blind color image watermarking scheme based on nonsubsampled contourlet transform (NSCT) and Schur decomposition. In the stage of watermark embedding, NSCT is applied to acquire low-frequency coefficients of host images, and non-overlapping coefficient blocks of sized 4 × 4 are selected from the layered low-frequency coefficients by using Hash pseudo-random algorithm based on MD5. The watermark information encrypted by Lorenz chaotic mapping is embedded into the maximum eigenvalue of the matrix by quantization index modulation. In the extraction process, watermark information is extracted from the attacked watermarked image without using any original information. The simulation results show that all values of peak signal-to-noise ratio of the proposed watermarking scheme are more than 40 dB under the payload of 1/16 bit/pixel, and the values of Structural Similarity Index Metric are more than 0.97; when subjected to various common attacks, the mean of normalized cross-correlation is above 0.96, and the bit error rate is about 0.05; the key space is more than 2⁶²⁴; the running time is about 1.5 s. Compared with other schemes, on the premise of ensuring invisibility and embedding capacity, the proposed scheme has been significantly improved in terms of security and robustness.