This paper proposes a novel method for vector image segmentation as the first step in developing a content-based vector image retrieval system. Structure of a vector image can be represented as a tree in which each node is assigned each object region in the vector image and each link repre- sents the inclusion relation between two object regions. In order to generate such trees, the method separates object re- gions from background by detecting figures defining bound- ary between object regions and background. The proposed method finds object regions before rasterizing, which is an essential difference from existing object separation meth- ods. We have evaluated the effectiveness of the proposed object separation on 40 test vector images by comparing manual object separation. The experimental results have shown that the proposed method has a high performance which is comparable to manual object separation.