Artwork Identification method and database construction

For identification an image, there are some researches published. Through the experiments, we have learned that there are some limitations in achieving a high matching rate. In this paper, we propose an effective data construction for identification artworks to improve the identification rate and reliability. Some image identification methods based on feature points have the limitations like large-scale size conversion, geometric distortion, camera angle distortion and so on. For overcoming the limitation, we construct the image database with several images for an artwork. We attach labels to image. The images are made up of the original image file of artwork, size conversion of various scales, various kinds of geometrical distortion, and camera shots at multiple angles. To eliminate unnecessary feature inclusion, we remove the boundary of image that included information of non-artwork. We matching the feature and get the ranking with these database. Through analyzing the ranking, we can get the result of artwork identification finally. The experiments shows the better identification rates, especially in the high-scale attacks.