Digital image watermarking using Learning Vector Quantization and hyperanalytic wavelet packet transform

Digital watermarking techniques have been been largely applied into copyright protection and authentication of multimedia data. This paper proposes a novel digital watermarking algorithm for digital images by combining the hyperanalytic wavelet packet transform (HWPT) decomposition and Learning Vector Quantization (LVQ) neural network. For inserting watermark, the non-overlapping blocked original images are decomposed according to two-dimensional hyperanalytic quadrantal wavelet packet and the preprocessed watermark image is embedded into the selected coefficients windows with different angles. Subsequently, in order to accomplish watermark strength maximum and to decrease the visual distortion, a competitive learning procedure is applied to train with the set of LVQ training patterns and the trained LVQ is attributed automatically to classify a set of testing patterns while encoding corrected errors. Thus, it maintains the visual quality and and meanwhile reduces the error rate by providing maximum possible required information. The simulation results demonstrate the proposed watermarking procedure has remarkable performances in the imperceptibility and robustness to general signal processing operations and some geometric attacks.