Logo Recognition Using Textual and Visual Search

Abstract

The amount of digital data transmitting via interne t has reached an enormous level. In order to conduc t efficient web data analysis, effective web mining t ools are needed. Logos, which represent companies’ brands, are highly regarded in a business world. These logos embedded in ordinary pictures could giv e an indication of popularity of the companies and their products in a region. Therefore, it is imperative to build a computer system to extract company logos from the se pictures. In this paper, a Logo on Map (LoM) system is proposed, which consists of three modules : picture extraction module (PEM), logo matching module (LMM) and web mapping module (WMM). Only the first two modules are covered in this paper. The PEM is based on a keyword textual search while the LMM is a visual search using SIFT (Scale- Invariant Feature Transform) algorithm. The three e xperiments are conducted using different sets of pictures extracted from the Flickr® website. The ex perimental results have proven that visual search i s more accurate than textual search and also demonstr ated that LoM could be used to discover hidden knowledge beyond logos.