Logo Detection Using Painting Based Representation and Probability Features

In this paper, a coarse-to-fine logo detection scheme for document images is proposed. At the coarse level of the proposed scheme, content of a document image is pruned utilizing a decision tree and a small number of features such as frequency probability (FP), Gaussian probability (GP), height, width, and average density computed for patches. The patches are extracted employing the piece-wise painting algorithm (PPA) used for text-line segmentation. The fine level of the proposed scheme refines the detection results by integrating shape context descriptors and a Nearest Neighbor (NN) classifier. We evaluated the proposed approach using a public and two large industrial datasets. From the experiment on Tobacco-800 dataset, the best precision and accuracy of 75.25% and 91.50% were obtained respectively.