In this paper, we propose a unified face anonymization framework that integrates attribute-driven identity separation and spatial mask optimization, specifically designed to achieve an optimal balance between identity privacy protection and attribute utility preservation. Existing face anonymization methods are limited by insufficient disentanglement of identity and attribute features and lack of fine-grained control during image synthesis, making it challenging to simultaneously ensure privacy and attribute fidelity. To address these challenges, our framework consists of two main modules. Specifically, an attribute-guided separation mechanism is employed to achieve effective disentanglement of identity and attribute features, while differential privacy-based perturbation is used to enhance the irreversibility and security of identity information. Additionally, spatial region masks are introduced to guide the fine-grained optimization of latent representations, enabling precise and coordinated control over identity, attribute, and background regions during image synthesis. Experimental results show that the proposed method achieves better performance than existing approaches in terms of identity obfuscation, attribute preservation, and image quality, significantly enhancing the utility and robustness of anonymized faces.