A New High Robust, Hybrid, Adaptive, and Blined Digital Image Watermarking Based on Machine Learning Using Joint DCT-DWT abstract
In this paper, a digital watermarking plan has been implemented utilizing Adjacency Extreme Learning Machine (AELM) on images and results are compared to other existing strategies. The adjacency relationships among the pixel in image can be utilized as reference features. AELM is utilized as a regressor . Digital watermarking issue can be treated as regression issue can be trained at the embedding methodology and watermark or logo or sequence can be embedded. The use of Discrete Cosine Transform (DCT) in view of 8×8 sub-blocks to transform the image from spatial to the frequency domain. The
DCT results will make AC and DC coefficients. Then, at that point, the DC coefficients are assembled into a matrix to be transformed with Discrete Wavelet Transform (DWT), then the matrix splits to 3×3 overlapped sub matrices, the center DWT coefficient of every sub matrix, utilized for watermarking, considering of human vision system (HVS), empirical, and structural adjacencies on its adjacent coefficients. Our extracting process is visually blind, it implies there is no requirement for original image for extracting the watermark bits. In the AELM algorithm, DWT coefficients consider as input for AELM network, for extracting HVS, and adjacency dependency features of them, that are vital for robust watermarking. The AELM is exceptionally quick, cost delicate and has great learning and speculation capacity, the watermark can be accurately extracted according to the watermarked image attacked by the pernicious attacks. Experimental results show that the AELM based watermarking plan outflanked other existing strategies against various attacks. As implemented digital watermarking plan is robust and subtle decided in light of determined measurements PSNR, and SIM.
