Fast retrieval on large scale human face data sets is a key problem in face recognition applications. The face Hashing method with short length can reduce the computational amount of face feature alignment, and is helpful to the application of large-scale face recognition. In this paper, a deep face Hashing method based on the loss function of ternary-group is presented. By optimizing the loss function of ternary-group, the deep convolution neural network is trained to extract the deep feature of images. The distance between the similar images can be as small as possible, and that between different kinds of images is as large as possible. The high dimension feature is mapped to the low dimension space by adding the random mapping layer following the deep network. And the low dimension space is further mapped to Hamming space by the threshold algorithm. Experimental results on multiple standard datasets show that the proposed method outperforms other state-of-the-art methods.