A Novel Feature Extraction Dual DCT-DWT Image Watermarking Combined with Chaos-Based Cryptosystem

Abstract

As a result of the emerging technologies such as Internet-of-Things (IoT), data analysis and automation in many aspects of identity, watermarking has become increasingly important. A majority of prior systems focused on watermark embedding and recovery techniques for binary and grey-scale images in one specified region of the cover image, making the system less dependable. This study develops a novel feature extraction based blind cryptographic watermarking mechanism which provides multi-layered protection to digital data to significantly improve the watermarking system's characteristics such as robustness, visible quality and security. A hybrid algorithm is proposed by combining DCT and multilevel DWT transformation technique to provide copyright protection and authentication to digital image over the internet. Unlike the previous model here both the host and watermark images are considered colored. Prior to embedding, a layer wise encryption is performed on watermark using Chaos-based cryptosystem to enhance the security of proposed system. It is processed by applying confusion and diffusion process on individual RBG (Red, Green and Blue) components of watermark image. Then the cover image and watermark image are directed by DCT coefficient and multi-region Discrete wavelet transformation to embed the watermark in both high and low sub band of cover image. Thus, by embedding the watermark twice in low and high regions of the cover image, the algorithm's robustness is considerably improved. For better imperceptibility, the proposed method also applies novel perception using SSIM_lPT metric to extract the similar features between cover image and watermark image. This is done by dividing the cover image and watermark image into non overlapping blocks of size 8x8 respectively. Now, the best match blocks of cover image are chosen for further embedding a watermark in both low-frequency and high-frequency regions of cover image. In addition, the use of non-consecutive blocks of pixels to contain the watermark makes the system more resistant to a range of attacks. In an experimental scenario, the approach is evaluated using a variety of quality criteria and watermark removal attacks. In comparison to other existing systems, the results suggest that the model can allow invisible watermarking as well as excellent attack resistance.