Digital image watermarking based on regularized filter

Abstract

This paper presents a spatial domain image watermarking method based on regularized filter. In the proposed method, a watermark image is embedded into a host color image directly by modifying the blue color component. The watermarkmark strength is controlled by two factors, i.e. a constant value and the luminance within a local embedding area. The prediction of the original host image is obtained from the watermarked image by using the regularized filter, so that the embedded watermark can be blindly recovered by substracting the predicted image from the watermarked image. Set of experiment are carried out to verify our proposed method. The results show that the accuracy of the extracted watermark in terms of NC was higher, compared to the previous method. The robustness comparison between both methods against different types of attacks is also performed and presented.

Adaptive Optimized Discriminative Learning based Image Deblurring using Deep CNN

“…To attain PSF of the blur image as uniform, Kumar et al [34] employed neural networks under learning-based techniques for deblurring process. Linear-Time-invariant-Regularized-Backward-Heat-Diffusion (LTI-RBHD) developed by Wang et al [35] estimates blur kernel with different widths which results in better performance compared to wiener filtering method. An iterative method was developed by Leon Lucy and William Richardson to reconstruct the latent image from blurred image with known power spectral density.…”

“…A Nash equilibrium method is used to obtain decoupled iterative BM3D deblurring method known as IDDBM3D in [36]. Gong et al [37], Sun et al [38] and Zhang et al [39] developed deep learning methods based on CNN and RNNs.…”