Abstract
3D mesh watermarking in the transform domain requires significant computational complexity. This is due mainly to the incessant use of high-resolution meshes which require more and more resources. Normally, this is an expensive work that harms the commercial chain of low computational cost applications requiring content protection or enrichment. To tackle this issue, we proposed herein a high-capacity and blind watermarking scheme for 3D multiresolution semi-regular meshes while maintaining a trade-off between efficiency and robustness. For this purpose, our solution uses an unlifted butterfly wavelet transform technique that explores the computing power of the Graphic Processing Units (GPU) architecture and the Open Computing Language (OpenCL) framework. The robustness was optimized by generating a turbo-encoded watermark. This latter is embedded in the wavelet coefficients after their spherical parametrization at various levels of details using the least significant bit technique. The method allows a better imperceptibility of the watermark and invariability to affine transformation. It also shows comparative robustness against most of the geometric attacks including additive noise, quantization, smoothing and compression. Moreover, the comparison with other serial watermarking schemes proves the effectiveness in terms of computational complexity of our method. OpenCL embedding implementation offers 3-9 x speedups with a low-power GPU architecture for different mesh sizes. In case of extraction procedure, the speedups obtained vary between 2 x and 12 x.