In order to avoid the details blurring and noise amplification, the image is decomposed as structural layer and texture layer, and the dehazing operation is only performed on the structural layer. The transmission fusion method based on the idea of frequency domain filtering is proposed to remove the block effects in the transmission image and the halo artifacts in the restored image. To solve the problems existed in the transmission optimization process such as the complex computation, being incapable of keeping the balance between the transmission smoothing and details preservation, we propose the multi-guided filtering method. At the same time, an adaptive atmospheric light estimation method is proposed which can be applicable to the scenes with large white objects. Experimental results show that the proposed algorithm can remove the haze effectively and the restored image has clear details and natural color. The noise and halo artifacts are suppressed remarkably, and the contrast and saturation of the scene are improved significantly.