A visible watermark conveys ownership information, used to identify the originator of an image. The watermark image is preprocessed and visibly embedded with an original image, before posting on the web for free observation and download. Based on the request of the buyer, the watermark can be removed to restore the unmarked image. A user-key-dependent frequency domain removable visible watermark embedding system (RVWES) is proposed. The user key structure decides both the embedded subset of the watermark and the host information adopted for adaptive embedding. With the correct user key, watermark removal can be carried out for information detection, and the high quality unmarked image can be restored. The proposed approach is mainly designed for the watermarking of digital camera images of various sizes. Pipeline and parallelism mechanisms are used to improve the throughput in the RVWES. The hardware implementation algorithm is tested in a Xilinx Virtex V XC5VlX330 technology based field programmable gate array, with which the throughput achieved is 4.3 Gbit/s with a total area utilization of 17448 BELs. To improve the robustness, a chaotic map is used to shuffle the watermark. The experimental results show that the watermarked image quality is good. Furthermore, the proposed approach is able to resist against a variety of general image processing attacks.