This paper proposes a dual watermarking algorithm for protecting the original feature information and carrying watermark information. One of the dual watermarks is image feature, and the other is the logo watermark. Firstly, Curvelet transformation is used to extract images texture features, which is considered as a feature watermark. Then, in the improved YCbCr color space Y$'Cb'Cr'$,$Y'$ component is divided into blocks, which is performed DCT transformation and SVD transformation in turn, and the feature watermark is embedded on the maximum singular value of each block by quantization index. Cb$'$ and Cr$'$ components are performed Curvelet transformation, their coefficients of low frequency are analyzed, and the second watermark is adaptively embedded according to the energy distribution. Futhermore, an improved space map is obtained by inverting the frequency coefficients of the three channels embedded in the watermark and converting the graph into the RGB space to obtain the color image with double watermarks. Finally, the performance of the proposed algorithm is verified, and compared with related algorithms through scientific experiment methods, the experimental results show that the double watermark embedded image has good invisibility and robustness, and the watermark of characteristics has an effective protection of the original image texture feature.