سیستم جدید برای تقسیم بندی و بازیابی علائم تجاری

With the increase in the number of trademarks, trademark imitation has become a serious problem. Thus, building an efficient trademark retrieval system is imperative. In this paper, such a system is presented. First, a semi-automatic segmentation method is proposed to extract the shapes of those representative objects, called ‘masks’, in each trademark. Next, some features are selected to describe a mask. These include invariant moments, the histogram of edge directions, and two kinds of transform coefficients that are robust to geometric deformation. Then, based on the rank of the feature distance, a similarity measure is provided to do the similar trademark retrieval. Finally, a feedback algorithm is also proposed to automatically determine the weight of each feature according to the user's response. Furthermore, in order to show the effectiveness of the proposed system, two databases from MPEG-7 test database are used to compare the performances of the proposed system and those methods using chain code, Zernike moments or MPLV as features. The experimental results show that the proposed system is superior to others.

Trademarks are specially designed marks that identify companies, products, and services. The imitation of a registered trademark is illegal. However, there are so many trademarks around the world and how to avoid designing a trademark similar to an existing one becomes an important problem. To treat this problem, developing an automatic and fast content-based trademark retrieval system is necessary. In general, trademarks can be divided into three types: character-in-mark, device-mark and composite-mark (Fig. 1). Since a character-in-mark trademark contains only characters, the traditional character recognition techniques can be applied. The device-mark trademarks contain only geometric shapes, while the composite-mark ones contain both characters and geometric shapes. The QBIC system proposed by IBM [1], [2] and [3] places more emphasis on the device-mark and composite-mark trademarks. However, the QBIC system does not work very well for some trademarks with geometric deformation or partial change. In addition, if the users do not satisfy the retrieval results, the QBIC system cannot take the user's response to retrieve again.

In this paper, an efficient and effective trademark retrieval system is proposed. In the beginning, we have proposed a user-friendly and semi-automatic trademark segmentation sub-system to extract the desired masks from a trademark image. Then, four features are presented. Invariant moments are used to treat rotation, scaling and translation invariant. Lap and Der features are robust to geometric deformation with similar mass center. The histogram of edge directions is especially effective for searching polygonal masks with different mass centers. In the retrieval process, a grade evaluation method is provided to measure the similarity between a query mask and each matching mask. Finally, we have introduced a feedback algorithm to automatically and interactively determine the weights of features according to the user's response. Based on the feedback algorithm, the advantage of each feature can be enhanced. The experimental results show that the proposed system has a better performance than the other methods. In addition, our system is also efficient in time complexity. For example, the time of extracting all the proposed features for D1-A is about 20 min by using the Pentium III 800. However, to calculate the MPLV for D1-A needs about 4 h.