In order to solve the problem that the existing polarization algorithms relied on the sky region to estimate the atmospheric parameters is interfered by white targets or highlighted regions, a universal multi-scale polarization dehazing method for image is proposed. The quadratic tree spatial index and image dark channel prior method of polarization difference image are researched, which break through the limitations of the estimated model parameters depending on the sky to reconstruct the scene depth, and restore low-frequency foggless images with atmospheric scattering model. At the same time, the soft threshold denoising algorithm is studied to solve the residual noise problem in the target restoration process, combining with the transmission rate reconstructed by low-frequency information, the texture details are enriched by gradient enhancement. Finally, the clear image is reconstruct through wavelet. The experiment results show that the proposed algorithm effectively eliminates the limitation of estimating the atmospheric parameters subjected to the sky region, and suppresses the influence of noise with the operation efficiency improved greatly. The target in restored image is more clearly, and the details are more abundant.