Sketch Retrieval Based on Spatial Relations

Shuang Liang, Zhengxing Sun, Bin Li

Abstract: In this paper, we propose an effective approach for sketch retrieval. Given sketches are firstly decomposed into basic geometric primitives and represented as a topological graph that encodes both the intrinsic attributes of the primitives and their relationships. The graph spectrum descriptor is then adopted to translate the graph-match into the computation of vector distances. Sketches retrieval is finally done by means of similarity calculation of feature vectors based on dimensionality reduction. The relevance feedback is also introduced to refine the retrieval results. Experiments prove the effectiveness and efficiency of the proposed method in sketch content retrieval and user independent.