Low complexity template-based watermarking with neural networks and various embedding templates

Color image copyright protection based on watermarking using machine learning in the wavelet-packets domain

“…Again, the output imperceptibility was around 44 dB. Dzhanashia and Evsutin [30] proposed a low-complex template-based robust watermarking technique that utilized the neural networks during the extraction process. The scheme introduced a variable embedding strength α to control the imperceptibility against various attacks.…”

“…To aid the performance of robust image watermarking, modern classifiers were, recently, tested. This includes models like Particle Swarm Optimization (PSO) [14], Support Vector Machines (SVM) [15], Random Forest (RF) [16], and Artificial Neural Networks (ANNs) and deep learning approaches [17][18][19][20][21][22][23][24][25][26][27][28][29][30][31]. Detailed survey on these classifiers has shown that: a) PSO exhibited relatively high robustness with poor performance against cropping and JPEG compression; b) SVM experienced complications especially when choosing the proper kernel, and when applying the finetune parameter regularization; c) RF has revealed difficulties when training untested data with increasing complexity raised at increased data size, and, d) ANN exhibited better performance over PSO, SVM and RF, since it can be easily learned, even with complex patterns and large dataset sizes.…”