The presence of haze will significantly reduce the quality of images, such as resulting in lower contrast and blurry details. This paper proposes a novel end-to-end dehazing method, called Encoder and Decoder Dehaze Network (ED-Dehaze Net), which contains a Generator and a Discriminator. In particular, the Generator uses an Encoder-Decoder structure to effectively extract the texture and semantic features of hazy images. Between the Encoder and Decoder we use Multi-Scale Convolution Block (MSCB) to enhance the process of feature extraction. The proposed ED-Dehaze Net is trained by combining Adversarial Loss, Perceptual Loss and Smooth L1 Loss. Quantitative and qualitative experimental results showed that our method can obtain the state-of-the-art dehazing performance.