Nowadays, security and legal applications both heavily rely on surveillance cameras. However, using various video editing software, the photos and video recordings can be easily edited. The captured information can be used as evidence of crime scene investigation. The integrity of image and video can have a significant impact on the outcome of a legal case or investigation. Therefore, it's critical to confirm the authenticity of surveillance photos and videos before using them as proof in any legal situation. Existing deep learningbased approaches for picture copy-move forgery detection are ineffective in identifying the boundaries of small manipulated objects because they do not efficiently leverage high-resolution encoded features. The recovered frames in this proposed article utilizes the spatial information present in the feature maps and increases the precision of identification for small objects. The paper presents a customized CNN -LSTM layer that uses the transfer learning to distinguish between genuine and altered frames. The model evaluation is done using Ensemble Learning. By incorporating advanced neural network architectures, the model can effectively learn and extract complex features from both videos and images, enhancing its ability to detect copy move forgery. The model produces an accuracy of 95.00% for identifying forged videos of static background, 98.67% for identifying forged videos with moving background and 95.08% for identifying and localizing the forged Image. The model is tested against various social media images and videos. The experimental results demonstrate that the suggested model surpasses the performance than other existing approaches.