A Blind Watermarking Scheme Using Adaptive Neuro-Fuzzy Inference System Optimized by BP Network and LS Learning Model

Abstract

To maintain a trade-off between robustness and imperceptibility, as well as secure transmission of digital images over communication channels, a digital image blind watermarking scheme on the basis of adaptive neuro-fuzzy inference system (ANFIS) is proposed in this study. To achieve better results, an optimized ANFIS (OANFIS) combines a back propagation neural network and a least-square (LS) hybrid learning model. Each 3 × 3 non-overlap block of the original host image is selected to form a sample dataset, which is trained to establish a model of nonlinear mapping between the input and the output. As a sequence, the host image is decomposed by wavelet transform to obtain a low frequency subband. Finally, each watermark signal is adaptively embedded into the low frequency subband by OANFIS. Results show that our scheme is robust to various attacks while effectively maintaining satisfying transparency.