Abstract: Image annotation is an active research topic in recent years.In this paper,we targeted at solving the automatic image annotation problem in a novel search and refinement framework.In the search stage,we performed content-based image retrieval(CBIR) based on the MSF global feature to find similar images from image database.Then in the refinement stage,an algorithm using random walk with restarts is used to re-rank the annotations.The refinement keywords are used to annotate the uncaptioned image.This framework does not impose an training stage,but efficiently utilizes well-annotated images,and is potentially capable of dealing with unlimited vocabulary.The experiment results show the effectiveness and efficiency of the proposed approach.

Key words: Image annotation,Similarity search,Refinement,Markov,Content-based