راه حل موثر برای بازیابی تصویر علامت تجاری با استفاده از ترکیب توصیف شکل و تطبیق ویژگی

Trademark image retrieval (TIR), a branch of content-based image retrieval (CBIR), is playing an important role in multimedia information retrieval. This paper proposes an effective solution for TIR by combining shape description and feature matching. We first present an effective shape description method which includes two shape descriptors. Second, we propose an effective feature matching strategy to compute the dissimilarity value between the feature vectors extracted from images. Finally, we combine the shape description method and the feature matching strategy to realize our solution. We conduct a large number of experiments on a standard image set to evaluate our solution and the existing solutions. By comparison of their experimental results, we can see that the proposed solution outperforms existing solutions for the widely used performance metrics.

With the rapid increases of multimedia information, multimedia information retrieval, e.g., content-based image retrieval (CBIR) [1] and [2], has drawn more and more attention. There is a growing interest on CBIR from both academia and industry. As a branch of CBIR, the research of trademark image retrieval (TIR) is of great practical significance. For example, if a company want to register a new trademark, they must find whether there are any similar trademarks in existing database to avoid trademark infringement. Existing trademarks retrieval is mainly based on manual classification code [3]. With the increase of registered trademarks, finding similar trademarks by human becomes laborious; thus, it is of great importance to find effective solutions for TIR [4].For realization of TIR, two important issues must be addressed. One is how to extract appropriate feature vectors to represent image content correctly, and the other is how to carry out the image retrieval based on the extracted feature vectors effectively. For trademark images, the shape feature vectors are usually used to represent image content, so in this paper we concentrate on shape-based solution for TIR. Jain and Vailaya [5] proposed the weight-based solution (WBS), in which the extracted feature vector includes two shape features: edge directions and invariant moments, and a weight-based strategy was presented for feature matching. Wei et al. [6] proposed the two-component solution (TCS), in which centroid distances, contour curvature and Zernike moments were selected as the shape features, while a two-component strategy was applied in feature matching. Now, let us look at a simple example to investigate the performance of these two solutions.

In this paper, we address the problem of trademark image retrieval (TIR) by proposing a novel solution which consists of an effective shape description method and an effective feature matching strategy. In the shape description method, two feature descriptors are presented. The contour-based feature descriptor not only includes the histogram of centroid distances, but also represents the relationship among two adjacent boundary points and the centroid. The region-based feature descriptor includes the feature points matching and the spatial distribution of feature points. In the feature matching strategy, a statistics-based method was proposed to compute the dissimilarity values between shape feature vectors of images. Finally, we conduct a large number of experiments based on the standard image set to evaluate the performance of our solution. The experimental results show that our solution outperforms two existing solutions for the widely used performance metrics. In future work, we will consider how to further improve the robustness of our shape descriptors in order to apply our solution to other applications in CBIR.