Image forgery localization is utilized to identify areas of a digital image that have been manipulated while ensuring the image's authenticity. Currently, deep learning-based techniques have been extensively employed in image forgery detection and localization with notable achievements. However, contemporary deep learning techniques utilize image content and high-frequency data as inputs. High-level features (for instance, brightness inconsistency) and low-level features (such as camera fingerprints) are separately extracted, then combined at the end of the network. This leads to a lack of exchange of information and guidance between the two feature types during the extraction process, inhibiting the network's ability to improve recognition accuracy in a complementary manner. Therefore, in this paper, we propose Dual-stream Feature Exchange Enhanced Network (DFEE-Net), in which low-level features guide the extraction of high-level features in the encoding stage, while in the decoding stage, the two streams guide each other to extract useful features through information exchange. Experimental results support that the interaction of information enhances the network's ability to recognize tampered regions with improved accuracy.