A Human Visual System-Based Blind Image Watermarking Scheme in the IWT-APDCBT Domain Integrating Geometric and Non-Geometric Distortion Correction Techniques

This work presents a robust digital image watermarking scheme to protect digital images from illegal usage and manipulations. The excellent re-constructional attributes of integer wavelets transform (IWT) coefficients are fused with all phase discrete cosine bi-orthogonal transform (APDCBT), which helps preserve the image's visual quality. Human visual system (HVS) parameters help to choose an optimal watermarking region to improve image imperceptibility to a greater extent. To achieve better watermark robustness in diverse attack conditions (in the presence of geometric and non-geometric attacks), we proposed two stages distortion correction technique for restoring watermarked geometry followed by a non-geometric correction. Geometric distortion correction (GDC) techniques have been implemented in a deep framework employing a pre-trained network of MobileNet-V2, customized to perform the task of a regressor to handle diverse geometric attacks. Statistical non-geometric correction technique addresses the non-geometric attack and restores the deviated coefficient values without considering any other local characteristics of the watermarked image. This method provides adequate image imperceptibility as well as satisfactory robustness of the watermark under both attack conditions. Also, the technique can give an average imperceptibility (in PSNR) of similar to 49.03 dB with a payload of 1024 bits. The experimental results suggest that the proposed scheme is secure and provides enhanced performance compared to existing techniques.