At a time when there are numerous cases of digital misinformation and advanced modes of image manipulation, it is very important to verify the authenticity of the digital media today in sectors such as cyber security, journalism, law enforcement and digital forensics. In this work, a transdisciplinary auto image integrity verification system is developed that integrates the extensive CNN features with the metadata-based forensic understanding, supporting information forensic, artificial intelligence, and even metadata science. The main goal will be the increase of tamper detection and provenance confirmation using multimodal analysis pipeline. The proposed system combines features defined by placing a spatial domain using the ResNet-50 and EfficientNet structures with metadata entropy measures specific to the exact size of Exif operands, GPS consistency tests, and anomalies in time stamps. The joint embedding model and a bimodal attention mechanism are used to fuse the information on a feature level and make the interpretations mutually reinforcing in regards to visual features and metadata information. It was trained and validated on two benchmarks CASIA v2.0 and PSBattles on synthetic manipulations of splicing, copy-move and content-aware fill. In experimental results, the integrated model reached a F1-score of 94.2% with an AUC rating of 0.961 that is greater than that of the unimodal models by a factor of more than 7% in terms of classification performance. All metadata inconsistencies indicated more than 83% of visually valid tampered images, indicating the importance of metadata in validating forensically. This work highlights the efficacy of a transdisciplinary approach that combines the practice of deep learning with digital forensics and semantic metadata analysis to create the scalable solution to image integrity verification in the real-life scenario.