In recent years, automatic coloring of anime line drawings has attracted significant attention, with growing demand for intelligent coloring techniques. Traditional manual coloring methods are not only inefficient but also struggle to maintain color consistency, while existing deep learning-based approaches still face substantial challenges in establishing semantic correspondence between line drawings and reference images. To address this critical issue, this study proposes an innovative cross-attention based coloring method (Att-color) for anime line drawings. The core innovation lies in the design of an efficient cross-attention module that precisely captures multi-level semantic relationships between line art and reference images, thereby generating colorized results with high semantic consistency. Systematic experimental validation demonstrates that our method significantly outperforms existing techniques in semantic correspondence, particularly in handling complex poses and richly detailed anime line drawings, achieving more accurate and stable semantic alignment.