| Summary: | 碩士 === 國立臺灣海洋大學 === 電機工程學系 === 100 === Registering a trademark is necessary for the protection of the trademark owner’s legal rights. Conversely, when registering a trademark, one must be very careful not to violate the rights of exiting trademarks owners. Therefore, it is highly desirable to be able to search and retrieve from the database all trademarks that may appear similar to the one about to be registered.
Attempting to fulfill the goal stated above, this research first uses the Scale Invariant Feature Transformation (SIFT) algorithm to extract local features of trademark images, then the Bag of Features (BoF) model to obtain vocabulary-based representation of the images. Finally, the Locality Sensitive Hashing (LSH) is applied to facilitate efficient indexing and retrieval of the images. The users of our system can search the database to retrieve resembling trademarks for further examination, and they can also add new trademark images into the database if so desire.
Because the amount of trademark images will only continue to grow, for future scalability of the system and to gain experience on cloud computing, we choose to deploy our database on Dot-Cloud, an IaaS-style cloud system. The database was implemented with MySQL and the Web program was developed with Django.
|
|---|
