A Robust Watermarking Algorithm for Video Game Artwork Based on Pose Estimation Neural Network

Video game artworks serve a crucial role in video game development, the associated copyright protection remain largely unsolved. We propose a robust watermarking for protecting the copyright of video game artwork. Specifically, the watermarking algorithm uses pose estimation neural network to select pose joint points, and construct feature areas with the pose joint points as the center. Compared with existing local feature watermarking methods, the pose joint points are selected by the pose estimation neural network are much more evenly distributed, stable and orderly. Next non-maximum suppression (NMS) is used to select non-overlapping feature areas. Then discrete Wavelet transformation (DWT) performs on each feature area, and then dividing each sub-band into blocks. Finally, each block is decomposed by QR decomposition and select R matrix to embed watermarks. In the extraction procedure, the video game artwork is corrected using affine transformation before extraction. Experimental results indicate that the proposed algorithm outperforms the state-of-the-art methods in terms of imperceptiblity, robustness and computational cost. The algorithm is robust to common image attacks and geometric attacks, can effectively resist special attacks including text addition, tampering weapon, and cutting body parts of video game artwork.