بازیابی مبتنی بر محتوا از آرم و علائم تجاری در پایگاه داده های تصویر رنگی نامحدود با استفاده از رنگ اج گرادیان نمودار هیستوگرام همزمان

In this paper, we present an algorithm that extends the Color Edge Co-occurrence Histogram (CECH) object detection scheme on compound color objects, for the retrieval of logos and trademarks in unconstrained color image databases. We introduce more accurate information to the CECH, by virtue of incorporating color edge detection using vector order statistics. This produces a more accurate representation of edges in color images, as compared to the simple color pixel difference classification of edges seen with the CECH. Our proposed method is thus reliant on edge gradient information, and so we call it the Color Edge Gradient Co-occurrence Histogram (CEGCH). We also introduce a color quantization method based in the hue–saturation–value (HSV) color space, illustrating that it is a more suitable scheme of quantization for image retrieval, compared to the color quantization scheme introduced with the CECH. Experimental results demonstrate that the CEGCH and the HSV color quantization scheme is insensitive to scaling, rotation, and partial deformations, and outperforms the use of the CECH in image retrieval, with higher precision and recall. We also perform experiments on a closely related algorithm based on the Color Co-occurrence Histogram (CCH) and demonstrate that our algorithm is also superior in comparison, with higher precision and recall.

Content Based Image Retrieval (CBIR) is an area of signal and multimedia processing with many promising applications. CBIR is an active area of research with many years of attention, producing a high volume of research output, causing many subsets of CBIR to exist. Because of this, in this study, we concentrate on one subset of CBIR, which is on logo and trademark retrieval. Logo and trademark retrieval has received much attention in literature, due to its many promising commercial applications. It is seen as an increasingly vital tool for industry, commerce and trademark registration [1] and [2], and for sports entertainment [3] and [4]. For any image retrieval system, a suitable representation or feature for representing these objects should be chosen to facilitate a meaningful comparison of the input query to the images in the database. To choose this feature, the amount of distortion expected from the object is taken into account. As there are many different features to use when performing logo and trademark retrieval, one may suggest that a hybrid approach is necessary: combining more than one feature into a single feature descriptor. However, for the feature representation in logo and trademark retrieval, different feature representations ultimately have different goals in mind, and is fairly difficult to create a hybrid approach in combining more than one feature into a single descriptor. However, for selecting a feature representation of interest, there are a wide variety of approaches to solve this problem, each having their own assumptions on the properties of the logos and trademarks to detect or retrieve.

This paper addresses unconstrained color logo and trademark retrieval in image databases, extending the unconstrained color object detection work by Luo and Crandall. Luo and Crandall’s work has the possibility of misclassifying edges in their edge map; therefore, we generated with a more reliable edge map with color edge detection using vector order statistics. To create this edge map, a threshold must be defined. A global threshold is not possible for all images, so an adaptive thresholding scheme must be sought. As such, we extend the binary segmentation framework by Otsu to generate an adaptive color edge detector. We also introduce a color quantization scheme based in the HSV color space, and demonstrated that it is a more suitable mechanism for image retrieval in comparison to scheme proposed with the CECH and the CCH. The retrieval results illustrate that our system can accurately determine the location of the logo or trademark of interest in spite of partial deformation, and produces higher precision and recall in comparison to the CECH and the CCH. These tests were performed on an unconstrained color image database, which poses as a challenge, for the logos and trademarks to detect in this database are subject to many uncontrollable factors. These include factors such as lighting, deformation and occlusion. The results demonstrate significant improvements to the CECH and CCH object detection frameworks, with the amount of correctly delineated logo images being higher in comparison to these two. However, there are several aspects of our system that could be improved upon, which provide much future work in this area. Better retrieval results can be achieved by creating a more robust similarity measure, as opposed to histogram intersection, though the use of histogram intersection generated promising results. We are investigating pre-screening approaches, where images in the database not containing the correct spatial–color statistics to that of the input logo are not retrieved, ultimately reducing computation time by effectively narrowing the search to images possibly containing the logo of interest. Finally, we are investigating the detection of logos and trademarks under significant deformations, and detecting multiple instances of a logo or trademark in an unconstrained database image.