<p>This paper introduces a groundbreaking approach for enhancing the detection and localization of multiscale splicing forgeries in images, a concern growing with the widespread dissemination of manipulated images online. Traditional methods often focus on singular attributes or overlook the significance of multiscale information, making them less effective for identifying varied sizes of forged areas. This study utilizes the Residual Attention and Multilevel Feature Fusion Network (RAMFF-Net) technique for enhancing the multiscale features. This technique uses a combination of residual attention and the integration of multiscale local and global information to significantly improve detection precision. Initially, our method consolidates multilevel convolutional feature maps through an encoder to refine feature representations, thereby boosting the model's capability to pinpoint forged regions across different scales. Moreover, this technique also uses a multilevel feature fusion block (MFFB) to refine these features further, amplifying the relevance of areas pertinent to the task while diminishing irrelevant information. Additionally, a global attention block (GAB) is designed to grasp the extended relationships across various image sections, equipping the model to adeptly manage intricate forgery cases. Focused on the CASIA dataset, our approach has shown remarkable effectiveness, achieving an F1 score of 89.7%, which surpasses the performance of existing state-of-the-art methods. This specificity indicates a tailored and highly effective solution for the challenges posed by image forgery detection and localization within the CASIA dataset context.</p>