With the widespread use of image editing tools, image manipulation has become increasingly easy, raising significant concerns about content authenticity across various fields. As a result, image manipulation localization (IML) techniques have become crucial. Most current methods perform pixel-level tampering detection by learning discriminative features. However, these methods typically rely on single-modality features and shallow fusion strategies, failing to capture the deep correlations between features from different domains. To address this issue, we propose the Multi-Dimensional Cross-Domain Collaborative Network (MDC-Net). MDC-Net combines RGB, noise, texture, and edge features, using wavelets with dynamically weighted correlations to extract texture features and designing fusion modules to exploit cross-domain complementarity with noise fully. Additionally, we extract edge features through differentiated edge detection operators and attention-guided feature interaction mechanisms to enhance boundary information representation. Experimental results demonstrate that MDC-Net achieves superior and stable performance compared to existing methods across several public image manipulation datasets.