Recently, the state-of-the-art deep learning methods have demonstrated impressive performance in segmentation tasks. However, the success of these methods depends on a large amount of manually labeled masks, which are expensive and time-consuming to be collected. To reduce the dependence on full labeled samples, we propose a novel Consistent Perception Generative Adversarial Network (CPGAN) for semi-supervised stroke lesion segmentation. Specifically, we design a non-local operation named similarity connection module (SCM) to capture the information of multi-scale features. This module can selectively aggregate the features at each position by a weighted sum. Furthermore, an assistant network is constructed to encourage the discriminator to learn meaningful feature representations which are forgotten during training. The assistant network and the discriminator are used to jointly decide whether the segmentation results are real or fake. With the semi-supervised stroke lesion segmentation, we adopt a consistent perception strategy to enhance the effect of brain stroke lesion prediction for the unlabeled data. The CPGAN was evaluated on the Anatomical Tracings of Lesions After Stroke (ATLAS). The experimental results demonstrate that the proposed network achieves superior segmentation performance. In semi-supervised segmentation task, our method using only two-fifths of labeled samples outperforms some approaches using full labeled samples.