Logo based document image retrieval using singular value decomposition features

Growth in information technology has increased repository of document images in various organizations, digital libraries and government offices. Logo based document image retrieval provides an easier way of searching and browsing of these documents. Duty noise, large ink blobs and choosing appropriate features are major challenges to implement logo based retrieval of document images. This paper presents a novel method for logo based document image retrieval, when a query document containing logo is given as an input. Logo from the query document is extracted based on energy contributed by connected components of document image after using Singular Value Decomposition (SVD). For retrieval of documents we propose a method using two set of features (i) Derived Singular Value Decomposition (SVD) features of extracted logo and (ii) Singular values of singular matrix after applying SVD to an extracted logo. Canberra distance is used for matching features of query document with the features of database of documents. Proposed method is tested for publicly available database Tobacco-800 and results are compared with current state of art. Using derived features of SVD of an extracted logo we got 76% precision; however 84% precision is achieved for same set of documents using singular values of extracted logo.