A New Multi-scale Method for Islamic Star Pattern Images Indexing

We propose in this paper a multi-scale method for effective indexing of Islamic Star Pattern images (ISP). These decor are rotationally symmetric about its center and composed by multiple-level designs of different colors about its center. The proposed method is based on the following steps: First, we detect the center of rotation of the ISP image. Second, we decompose the image into blocks of fixed size centered on the detected center. This decomposition allows to capture characteristic features at different levels of the image. Every block is characterized by the symmetry information and the second order statistic. A feature vector is proposed to capture these features and similarity measure is given. We demonstrate performance of our approach on an ISP image databases and compare it to fractal dimension based approach. The experiments show the clear superiority of the new approach against his competitor