Logo recognition based on a novel pairwise classification approach

Logo recognition is an important task in the field of document image processing and retrieval. Successful recognition of logos facilitates automatic classification of source documents, which has been considered as a key strategy for document image analysis. From machine learning point of view, logo recognition may be considered as a multi-class classification problem. In this paper, a novel multi-class pairwise classification method is proposed and applied to logo recognition application. The proposed system takes the advantages of simplicity and speed of the nearest neighbor classification algorithm and the strength of other powerful binary classifiers to discriminate between two classes. The method is first validated on a set of UCI Machine Learning Repository datasets and then applied to the real machine vision problem. The experimental results show that the proposed technique not only achieves better classification accuracy, but also is computationally more efficient for tackling the classification problems which have large number of target classes.