Most of existing hashing methods for image retrieval problems assume all images are given at the beginning. However, in some image retrieval problems, images may arrive or be labeled in an online or streaming manner. Current online hashing methods are fully supervised which assume all images come with labels. However, in real world big data environments, it is infeasible to have a fully labeled image set. Therefore, we propose a semi-supervised online hashing method to fully utilize the small portion of labeled images and the large amount of unlabeled images. Moreover, several variants of the proposed method are also proposed in this paper. Three databases are used to simulate online data environments. Experimental results show that the proposed method outperforms the stat-of-art online hashing methods and other baseline methods.