Digital image watermarking has become an effective solution for copyright protection and multimedia authentication. It is a challenge to achieve an optimal balance between imperceptibility, robustness, and security. This paper proposes a blind image watermarking framework that integrates transformation techniques, Grey Wolf Optimization (GWO), and a lightweight convolutional neural network called DWatNet. The host image is transformed using multiple transformation domain. The binary watermark is scrambled using ArnTran to improve security. GWO is used to determine an optimal embedding scaling factor to optimize visual quality and watermark recovery performance. Furthermore, DWatNet is used for automated validation of watermark embedding during the optimization process. The performance of proposed framework is evaluated using benchmark datasets under various image processing and geometrical attacks. It presents high imperceptibility, robust watermark recovery, and reliable verification performance. Moreover, the effectiveness of the proposed framework is validated using various performance metrics such as average PSNR and SSIM of 46.72dB and 0.9941, respectively. Also, the computational time to achieve the task is 0.98s. Also, several methodologies such as payload capacity analysis, security analysis, ablation study, and Wilcoxon signed-rank statistical test are performed for model validation. The comparative analysis shows that the proposed framework provides an effective balance between robustness, security, and watermark verification, such that it is found effective for copyright protection and multimedia authentication applications.