To address challenges in mural outline generation, including inadequate feature extraction, poor capture of edge details, and limited robustness, this study proposes an efficient method for accurately extraction outlines from mural images with complex textures and details. A new framework based on the Holistically-Nested Edge Detection (HED) model is developed innovatively by integrating a Scale Enhancement Module (SEM) and a Compressed Spatial Attention Module (CSAM). The SEM utilizes dilated convolutions to achieve multi-scale feature extraction, enabling enhanced capture of both local and global information and improving outline extraction accuracy. The CSAM module focuses on refined representation of local edge features, effectively reducing information loss during feature propagation. Additionally, a loss function based on the Dice coefficient is introduced to address class imbalance between edge and non-edge pixels, further improving edge detection performance. Quantitative and qualitative results demonstrate that the proposed method clearly outperforms in handling texture blur and local information loss, thereby has advantage in mural outline generation. It successfully improves the outcome of mural outline generation and validates the effectiveness of multi-scale feature fusion and spatial attention mechanisms in complex texture image processing. The proposed approach provides new insights and directions for the digital conservation and restoration of murals.