Coherent diffraction imaging (CDI) reconstructs the complex-valued object of a sample from measured diffraction intensities, but its phase retrieval performance is often affected by measurement noise and insufficient prior constraints. To address these problems, a guided-filter-enhanced alternating direction method (GFADM) is proposed. In this method, guided filtering is embedded into the ADM iterative framework as a structure-preserving prior, and a filtering step is introduced between the measurement projection and the object-domain support projection to improve noise suppression and artifact control. For samples with distinct absorption boundaries or pronounced amplitude structures, the amplitude estimate is first processed by self-guided filtering, and the filtered amplitude is then used as a guidance image to assist phase updating. This strategy helps improve the structural consistency between the reconstructed amplitude and phase. In addition, a piecewise decreasing filtering-strength schedule is adopted to balance denoising and detail preservation: stronger filtering is applied in the early iterations to suppress noise propagation and stabilize the reconstruction, whereas weaker filtering is used in the later iterations to retain fine structural details. Near-field and far-field CDI experiments demonstrate that the proposed method can effectively suppress noise propagation and artifact accumulation, reduce background fluctuations, and improve the distinguishability of local structures. Compared with the enhanced ADM (EADM), GFADM produces clearer line-pair features, higher line-pair contrast, and better structural fidelity. These results indicate that the proposed method achieves a good balance between noise suppression and detail preservation, and exhibits stable reconstruction performance under different imaging conditions.