Robust, Adaptive, and Blined Digital Image Watermarking Based on Exterme Learning Machine Using Joint DWT-DCT

In this paper, a multiple assessment criterion based blind watermarking scheme for grayscale image watermarking using Weighted Regularized Extreme Learning Machine (WRELM) is proposed. One-level Discrete Wavelet Transform (DWT) is applied on three standard test images. LL۱ sub-band coefficients are chosen for watermark embedding. WRELM is initially tuned with a fixed number of training data used in its initial phase and size of block of data learned by it in each step. The training set is developed by performing Discrete Cosine Transform (DCT) for LL۱ sub-band, previously divided to ۸*۸ coefficients DCT blocks. Nine maximum valued coefficients of every DCT blocks, selected in zigzag manner, consider as training set, which is fed to WRELM for training. Human Vision System (HVS) features, and entropy assessments of DCT coefficients are the our assessment criterions, product by WRELM training. The input vector of WRELM is an array with ۸ coefficients of training set (except the middle). The WRELM target is a vector related to input vector with equivalent ۸ values equal to middle coefficient of training set. The output of WRELM is predicted value for actual middle coefficient of training set correspond to our assessment criterions, it is very important for robust embedding the bit of binary watermark image, in this coefficient. Our extracting process is blind, it means there is no need for original image for extracting the watermark bits. Results show great visual nature of watermarked images and great comparability between the extracted and original watermarks from watermarked and attacked images. The PSNR and SIM are viewed as well optimized both in the event of watermarked and attacked images.