General regression neural network based image watermarking scheme using fractional DCT-II transform

A novel gray scale image watermarking scheme in frequency domain is proposed through the combination of image features, extracted using fractional discrete Cosine transform (DFrCT) with general regression neural network (GRNN). The watermark is a binary image which is embedded into the output obtained by trained GRNN based on the relationship between the low frequency (LF) DFrCT coefficient and its neighborhood of each selected block according to human visual system criteria. Due to better function approximation, learning and generalization capability of GRNN, extraction of watermark using trained neural network is quite successful. The transform order of fractional discrete cosine transform provides the security to the proposed scheme. Experimental results prove that the proposed image watermarking scheme is imperceptible as quantified by high peak signal to noise ratio (PSNR) and robust as measured by the bit correct ratio between the original watermark and extracted watermark.