Logo retrieval and document classification based on LBP features

Abstract

Logo is the strong entity for retrieval of Content-Based Information (CBI) from any complex document image. Logo is the primary and unique entity which is used to identify the ownership of the documents. Automatic logo detection and retrieval facilitates efficient identification of the source of the document and it is one of the interesting problems to the document retrieval community. Wes proposed a method based on Local Binary Pattern (LBP) for logo retrieval from document images. It is used to describe the logos both query and document logo. The candidate and query logos ares matched based on the cosine distance. Based on it, distance ranks are generated to estimate the relevance of the logo. Later, matched logos are retrieved at a selected threshold of 98%. The performance of the algorithm is experimentally validated and its efficiency is measured in terms of the mean precision at the rate 87.80%, and mean recall rate 88.20% as well as average F-measure 88.00%. © Springer Nature Singapore Pte Ltd 2019.