A hub image retrieval method and a device thereof for improving the retrieval accuracy of an image containing a hub. The method includes: performing feature extraction on a hub image to be processed containing a target hub, and obtaining a hub feature to be processed containing N sub-hub features, wherein the N sub-hub features at least include information characterizing the features of a hub window of the target hub; determining similarities between candidate hub features of each candidate hub image and the hub features to be processed; and selecting a candidate hub image matching the hub image to be processed from the plurality of candidate hub images based on the determined respective similarities.