[Purpose/significance] The purpose of this paper is to protect Dunhuang cultural heritage and provide a mobile visual search service for Dunhuang murals to help users obtain rich knowledge resources of Dunhuang murals efficiently and conveniently. [Method/process] In this paper, a mobile visual search model based on BoW-based imagelow-level feature matching and high-level semantic association based on topic tags is constructed. In this model, SIFTis used to extract the local features of the image, K-means is used to produce a visual dictionary composed of K visualwords, TF-IDF is used to calculate the vector of the image mapped to the visual dictionary, and the inner product is usedto calculate the similarity to match and sort the images. Then, this paper adds semantic tags based on the image topics,extracts the tags that best match the image for semantic association search, and collects 1200 Dunhuang mural picturesto verify the validity of this model. [Result/conclusion] The results show that the BoW + SIFT model takes 163ms whenthe number of visual words is 1000, and the accuracy of image search with the image size in the range of 0.5-2.5 times ismore than 83.7%. What’s more, this model can effectively search images with semantic association of topics.