The rapid advancement of image editing technologies has significantly increased the difficulty of distinguishing authentic images from manipulated ones, posing serious challenges in digital forensics, social media, journalism, and legal investigations. Traditional image forgery detection techniques often rely on handcrafted features, which exhibit limited robustness against complex tampering operations such as copy-move, splicing, and post-processing manipulations. To address these limitations, this paper presents a hybrid deep feature fusion framework for image forgery detection using multiple lightweight deep learning models. The proposed approach employs fine-tuned lightweight architectures to extract complementary high-level representations from input images. These feature representations are integrated through a fusion strategy to capture both local texture inconsistencies and global semantic characteristics associated with forged regions. The fused feature vector is subsequently classified using a Support Vector Machine (SVM), enabling effective discrimination between authentic and tampered images while maintaining low computational complexity. The proposed framework is evaluated on benchmark image forgery datasets using standard performance metrics, including accuracy, precision, recall, and F1-score. Experimental results demonstrate that the hybrid feature fusion strategy consistently outperforms individual lightweight models and conventional handcrafted feature-based methods by improving detection accuracy and classification reliability. Owing to its lightweight architecture and efficient feature representation, the proposed framework offers a practical solution for real-time image forgery detection in resource-constrained environments.