Logo Localization and Recognition Based on Spatial Pyramid Matching with Color Proportion

In this paper, we propose a novel logo localization and recognition model in which we use the spatial pyramid matching with color proportion to train a score function and the Efficient Subwindow Search (ESS) to localize region and recognize specific logo. Our model has two advantages compared with other re- lated algorithms: (1) this model has a ��(�� 2 ) time complexity in localizing logo regions which is much faster than the traditional sliding window approaches that has a time complexity of ��(�� 4 ) which make it cannot apply to real world applica- tions because of the high computational cost; (2) by means of our well-designed spatial pyramid matching embedded by color proportion, we can achieve a higher recognition rate even with the linear kernel SVM. We test our model in a logo da- taset called FlickrLogos-32 which contains 32 different logo classes by download- ing them from Flickr. Experiments show that our model both have fast enough speed and high correct rate.