Multi-Modal Retrieval of Trademark Images Using Global Similarity
MASSACHUSETTS UNIV AMHERST CENTER FOR INTELLIGENT INFORMATION RETRIEVAL
Pagination or Media Count:
In this paper a system for multi-modal retrieval of trademark images is presented. Images are characterized and retrieved using associated text and visual appearance. A user initiates retrieval for similar trademarks by typing a text query. Subsequent searches can be performed by visual appearance or using both appearance and text information. Textual information associated with trademarks is searched using the INQUERY search engine. Images are searched visually using a method for global image similarity by appearance developed in this paper. Images are filtered with Gaussian derivatives and geometric features are computed from the filtered images. The geometric features used here are curvature and phase. Two images may be said to be similar if they have similar distributions of such features. Global similarity may, therefore, be deduced by comparing histograms of these features. This allows for rapid retrieval. The systems performance on a database of 2000 trademark images is shown. A trademark database obtained from the US Patent and Trademark Office containing 63000 design only trademark images and text is used to demonstrate scalability of the image search method and multi-modal retrieval.
- Information Science