<p>The emergence of deepfake technology has reshaped the landscape of multimedia manipulation, challenging the very foundations of truth and authenticity. Deepfake face synthesis has been exploited by malicious actors to impersonate individuals, fabricate evidence, and perpetrate various forms of fraud, eroding trust in digital interactions. The study explores frequency characteristics and authenticity utilizing datasets from generative models- PGGAN, StyleGAN, and Stable Diffusion. Representation Similarity Analysis (RSA) assesses the authenticity of synthesized face imagery against real facial images, highlighting their resemblance with authentic data and the necessity for efficient detection methods. A novel deepfake detection framework is proposed, backboned by Dual Residual Network (DRN) architecture inspired from Residual Networks. This framework extracts residual traces from deepfake content, enabling precise identification. High-frequency components significantly enhance the accuracy of detection models, with minimal misclassification rates, though on an extremely slight margin when compared to low pass filtered and raw images. The proposed DRN model demonstrated exceptional performance across the DFFD and DFF datasets, achieving Equal Error Rates of 0.04% for DFFD PGGAN, and 0.02% for DFFD StyleGAN as well as the Stable Diffusion track in DFF dataset. The study exhibited consistency in the latest generative technique of Stable Diffusion with a marginal AUC gain of 40.89%. Potential scope of the design extends its applicability beyond deepfake face forgery attacks to domains such as healthcare and insurance fraud, even encompassing a wide array of multimedia domains.</p>