Region-based CNN for Logo Detection
Yu Bao, Haojie Li, Xin Fan, Risheng Liu, Qi Jia · 2016
Logo detection has been extensively studied because of its applications in many fields. Most existing studies for logo detection are based on hand-designed local features which have certain limitations. Recently, Convolutional Neural Networks (CNN), especially the series of region-based CNN (R-CNN) have achieved great success in general object detection due to its power of feature learning and representation. There are also few work on logo detection based on R-CNN, showing its superiority. However, logo detection presents its own challenges compared to general object detection. In this paper, we aim to experimentally investigate the appropriate architecture and settings of R-CNN for logo detection, by comparing several popular frameworks of R-CNN and considering the characteristics of logo objects in images. We expect our results will be helpful for designing a state-of-the-art logo detection system using handy machine learning techniques.