Underwater optical images exhibit color distortion and hazy blurring effects due to complex changes in the underwater environment and the absorption and scattering characteristics of water. The traditional Transformer model pays equal attention to all regions of the processed image, which increases the number of models computations and easily leads to overfitting. To address these challenges, we designed a Transformer method based on an underwater light mask (LMT-UIE). The overall LMT-UIE we designed consists of two parts: the underwater light mask generation module (LGM) and the underwater light mask-based Transformer module (MTF). We designed the LGM to generate an underwater light mask to mitigate the effects of uneven underwater illumination. We design the MTF to utilize the underwater light mask for feature aggregation, which makes the Transformer focus more on the key information and enhances the model’s comprehension of the underwater image. Furthermore, we introduce parallel attention in the MTF to improve the Transformer’s local modeling ability as well as its generalization ability. The LMT-UIE method not only effectively mitigates the color distortion and blurring problems of underwater images but also has a stronger ability to extract and model key information, which significantly improves the performance improvement of underwater image enhancement tasks.