Abstract: Aiming at the degradation phenomena such as low contrast, much noise and texture loss in images acquired in low light environment, a low light image enhancement algorithm (MixFormer) fusing global and local features was proposed. Based on the Retinex theory, the algorithm model consisted of three subnets: the decomposition net, the reflection net, and the illumination net, which were designed for image decomposition, noise suppression, and contrast and detail enhancement respectively. A hybrid module combining convolution and Transformer was designed in the reflection net to extract global and local features. Through the wavelet attention module, the illumination net was employed to complementarily enhance high-frequency and lowfrequency features, remarkably improving detail recovery capabilities. Compared with existing methods, the proposed algorithm exhibits superior performance, achieving average improvements of 2.69 dB in PSNR, 0.0816 in SSIM, and 0.055 in LPIPS on the LOL v1 and LOL v2 datasets. The results show that MixFormer has substantial advantages in brightness, contrast, and restoration of edge details, and the visual effect is significantly improved.