Aiming at the uneven distribution of haze concentration and color imbalance in haze weather images, a block-based image enhancement algorithm is proposed, which combines dark channel prior algorithm with multi-scale retinex with color restore algorithm. The idea of the algorithm is to map the atmospheric scattering model linearly to the scale of the Retinex image model, and get a new multi-parameter model with both transmittance and illumination value in the atmospheric scattering model and the incident image in the Retinex image model. The new original image of removing atmospheric scattering image is obtained according to the model, and the incident image of the whole image is calculated by using different scales of guided filtering. Then, the final high-frequency detail image is obtained by combining the atmospheric scattering image and color restoration factor. The whole image is divided into several regions in view of the uneven distribution of haze concentration in the image and the dynamic truncation value of each regional block is calculated by using fusion algorithm. According to different dynamic truncation values, the high frequency details of the whole image can be dynamically adjusted to obtain multiple local optimal images. These images are equal-weighted at the pixel level, and finally the optimal image can be obtained which can retain the local details. The proposed method is experimentally compared with other existing methods on subjective visual effect and objective evaluation, the results show that the method can effectively solve the problems of non-uniform fog and color imbalance in images, and obviously improve the quality of the image after fog removal.