Translated Trademarks Retrieval using Color Autocorrelogram for Extracted Textual Parts

Introduction

Trademark recognition has received an intensive importance in the literature of content-based image indexing and retrieval, recent applications are proposed and dealing with logo recognition including: patent office registration and copyright property protection (Doermann et al., 1996), image documents classification in which logos can be used to reduce the semantic gap problem (Zeggari et al., 2010), (Bagheri et al., 2013). Logos can be used also for vehicle type recognition in intelligent transportation system techniques (Psyllos et al., 2010).

Significant former research related to logo images retrieval problem have been described in the last years, (Wang and Chen, 2002) have proposed a content-based color trademark retrieval system, the trademark feature was computed as hit statistics based on the edge pixels and the smallest circle that covers the trademark. (Cerri et al., 2006) have proposed a three distinct families of measuring functions for size functions as shape descriptors namely: distances from points, projections and jumps. (Bagheri et al., 2013) used Dempster-Shafer theory to fuse feature classifiers including Zernike moments, generic Fourier descriptors, and shape signature. In (Gori et al., 2003), the authors have proposed a modified learning algorithm derived from the classical back-propagation, they used a weighed norm depending on the gradient of the company logos corrupted by artificial noise. (Alajlan et al., 2006) have proposed a geometry-based retrieval system for multi-object images, a curvature tree is used to model both shape and topology of image objects including holes. The authors proposed a matching scheme which agrees with many findings in psychology about the human perception of multi-object images. (Wei et al., 2009) have proposed a trademark image retrieval system, they used Canny edge detector to extracts global and local features. The global features capture the gross essence of the shapes while the local features describe the interior details of the trademarks. (Liu et al., 2014) proposed a kernel-based l₂ norm regularized least square (RLS) algorithm, in which they combine Kernel technique with the l₂ norm RLS algorithm to enhance the performance of a vehicle logo recognition system.

The previous works related to the problem of logo recognition deals with the general problem of retrieving a list of the most similar trademarks resemblant to the query image on the database. Most of the previous works focus on the invariance of the query image to the local deformations such as partial occlusion, broken curves and local noise, or they are adapted to the global deformations such as scale, orientation and affine transformations. The problem addressed in the current work is the retrieval of logos representing one company in different languages. Translated logos of a same company are globally similar in terms of color distribution and non-textual shapes, but they have different interior parts representing the textual name of the company or their product in another language.

The rest of this paper is organized as follows: In section 2 a brief introduction to trademark recognition is presented. The next section surveys the proposed algorithm of translated logos recognition. In Section 4, to verify the performance of the proposed method, experimental results with real trademark database are analyzed and evaluated. Finally, section 5 concludes this paper.