<p>Image splicing forgery is often used to create fake news and propaganda, which can lead to significant negative effects on society. Therefore, accurately detecting spliced images and locating their spliced regions is crucial. However, most current image splicing forgery detection methods tend to focus on deep features, often overlooking the integration of multi-level features. Additionally, the interaction and adjustment of complementary information across different domains are not considered during the fusion of RGB and noise features. An image splicing localization network based on context-aware and cross-domain multi-scale fusion is proposed in this paper. This network consists of two branches: one extracts splicing forgery features directly from the RGB domain, while the other extracts features from the noise domain. Considering the interaction and adjustment between complementary information from different domains, this paper introduces a Cross-Domain Context-Aware Fusion Module to efficiently fuse these two types of features. To better capture hierarchical representations, the Cross-scale Adaptive Enhancement Module is employed to extract and enhance the image’s deep feature information using the attention mechanism. At the same time, by combining the low-level and high-level features, the network can analyze splicing forgery traces more comprehensively. Extensive experimental results show that the proposed method outperforms the state-of-the-art methods in splicing forgery detection and localization.</p>