Cross-modal alignment is a promising yet challenging task in multimodal learning. Existing methods typically assess it by measuring the cross-modal semantic similarity from both global and local perspectives. However, these methods often neglect their potential interdependence. Specifically, global matching methods suffer from the over-compression of local features, while local matching methods rarely consider the inherent spatial topology of image patches. To address these limitations, we propose MG-Net, a unified framework with two collaborative modules: Multi-View Differential Mixer (MDM) and Graph-Guided Structural Region Selector (GSRS). The MDM is designed to capture discriminative global representations. It generates a series of views by decomposing feature vectors through multi-order differential operations, and adaptively fuses them via a lightweight Mixture-of-Experts (MoE) network. Meanwhile, the GSRS organizes image patches as a spatial graph and employs text-guided contextual reasoning to select spatially coherent and semantically complete structural regions. Extensive experiments on the Flickr30K and MS-COCO benchmarks demonstrate that the proposed MG-Net outperforms state-of-the-art methods in most cases.

Image-text matching remains challenging in big data processing. Matching accuracy is influenced by various factors, including the correlation between images and texts, feature extraction and fusion. Although activation functions play a crucial role in image-text matching, their design has received limited attention. This paper proposes an image-text matching model that utilizes multi-scale feature fusion based on a piecewise polynomial activation function. On one hand, a feature correlation optimization method is proposed to minimize the distance between paired images and texts. This method introduces large-scale downsampling odd-even feature embeddings and mean downsampling feature embeddings. After feature enhancement using a self-attention module, the odd-even feature embeddings are corrected with large-scale mean downsampling features to improve their representational ability. Additionally, a new multi-scale feature fusion method is utilized to enhance the robustness of the feature enhancement algorithm. On the other hand, we propose PCPAF (Piecewise Cubic Polynomial Activation Function), which offers advantages such as low computational cost,C1continuity, and superior generalizability. The PCPAF significantly improves model accuracy compared to existing activation functions. By adjusting the parameters of PCPAF, different activation functions can be derived, thereby improving matching accuracy in other image-text matching models. Experimental results on the Flickr30k and MS-COCO datasets demonstrate that the proposed model outperforms state-of-the-art models in terms of overall performance.