Automatic logo detection from document image using HOG features

Document image analysis and processing has drawn the attention of many researchers due to its real-time applications in day-to-day life. Document database comprising of logo provides a good opportunity for an easier way of indexing, searching and retrieval of the documents. Logo detection is an essential need for the implementation of any logo-based document indexing or retrieval techniques. This paper aims to develop an efficient logo detection method for document images. The major steps employed in the developed system include preprocessing of the input document, finding the connected components and classification of these components into the logo and non-logo candidates. The preprocessing step employs a median filter and a unique procedure for the removal of clutter noise to reduce the false detection rate. Histogram of Oriented Gradient (HOG) features and an SVM classifier are used to identify the logo and non-logo candidates of the document. The presented system is evaluated using Tobacco 800 dataset and the results are compared with existing techniques. The results show an improvement of 5% in average logo detection rate with the proposed work.