Abstract: In unsupervised graph-based hashing for large-scale image retrieval, many efforts have been made to bridge the gap between the learned graph embedding and the corresponding binary codes.