Cmu Trademark Retrieval System

In traditional trademark retrieval systems, the trademarks are first annotated with keywords that are then used for retrieval. The Vienna Classification [1] is a well-known scheme to achieve this purpose and it constitutes a hierarchical system that proceeds from the general to the particular, dividing all figurative elements into categories, divisions and sections. The STAR system developed by Wu et al. [2] is an example trademark system that makes use of the Vienna Classification with user interaction. However, this process requires a lot of manual labor in order to assign keywords. Also, the annotation for the same trademark may not be consistent with different users. As a result, it is more desirable to allow the user to input a query by providing a rough sketch and then the system is smart enough to automatically extract features from this query sketch and use them for comparison in order to search for similar trademarks in the database. The TRADEMARK system proposed by Kato et al. [3] used the graphical features that consist of the spatial distribution, spatial frequency, local correlation and local contrast. The spatial distribution is computed as the pixel density for each 8×8 block while the local correlation and the local contrast are computed among 4×4 block. The spatial frequency is computed as the run-length distribution of black pixels after dividing the image into 4 rectangular blocks in the horizontal and the vertical direction. These block-based features are not good to describe lines or contours. In the ARTISAN project [4], the boundaries of the components of a trademark are extracted and analyzed. The distance measure is computed by a linear combination between the distance of whole-image shape feature vectors and the distance between each query component and the closest-matching component of each image in the database on the basis of shape, size and absolute position. In this system, each component is considered independently and the spatial relation between components within the same image is not considered. The Zernike and Pseudo-Zernike Moments are used in Kim and Kim's content-based trademark retrieval system [5]. Manmatha et al. [6] used the seven Hu moment invariants [7] as well as curvature and phase histogram for trademark retrieval. In Ciocca and Schettini's work [8] on trademark retrieval, Hu moment invariants, histogram of edge directions and wavelet coefficients are used as the features. Jain and Vailaya [9] also used Hu moment invariants and histogram of edge …