A Novel Graph-Based Image Annotation Refinement Algorithm
Zheng Liu · 2009
A novel graph-based approach to automatically refine image annotation is presented in this paper. Given an unannotated image, a set of candidate annotations is extracted by the existing image annotation method. Then, each candidate annotation is converted to vertex of a graph and the semantic similarity between two candidate annotations is used as edge weight. Next, a heuristics graph algorithm solving weighted MAX-CUT problem is used to prune the noisy annotations. Experimental results demonstrate the effectiveness of our image annotation refinement algorithm.