Logo Detection and Recognition Based on Classification

摘要

Online product frauds in the booming e-commerce market have become a major concern for market surveillants and commercial companies. The logo detection plays a crucial role in preventing the increasing online counterfeit trading attempts. In this paper, a novel method based on Random Forest classification with multi-type features is presented to detect the logo regions on arbitrary images and the detected logo regions are further recognized using the visual words with spatial correlated information. Extensive experiments have been conducted on realistic and noise images with different logos. The results show that the proposed method is able to detect the logo regions, and the recognition performance outperforms the well-known Viola-Jones approach for recognizing the arbitrary logos on realistic images.