Abstract: In order to improve the ability of resource integration and image analysis in digital libraries,
this paper proposes a content-based image retrieval method, which extracts SURF features of images as
visual words based on bag of features method and uses K-means method to build a visual dictionary. It
then obtains histogram features of images according to the mapping relationships between visual words
and visual dictionary. At last, the similarity between histograms is used for image retrieval. Practice shows
that the system has good operational efficiency and retrieval accuracy, and fully considers the illumination,
perspective, image size and other factors. It can not only promote and improve the working process in
library but also make corresponding technical improvement according to specific needs, which is suitable
for further promotion in digital libraries.