Fast and Effective Bag-of-Visual-Word Model to Pornographic Images Recognition Using the FREAK Descriptor

Recently, the Bag of Visual Word (BoVW) has gained enormous popularity between researchers to object recognition. Pornographic image recognition with respect to computational complexity, appropriate accuracy, and memory consumption is a major challenge in the applications with time constraints such as the internet pornography filtering. Most of the existing researches based on the Bow, using the very popular SIFT and SURF algorithms to description and match detected keypoints in the image. The main problem of these methods is high computational complexity due to constructing the high dimensional feature vectors. This research proposed a BoVW based model by adopting very fast and simple binary descriptor FREAK to speed-up pornographic recognition process. Meanwhile, the keypoints are detected in the ROI of images which improves the recognition speed due to eliminating many noise keypoints placed in the image background. Finally, in order to find the most representational visual-vocabulary, different vocabularies are generated from size 150 to 500 for BoVW. Compared with the similar works, the experimental results show that the proposed model has gained remarkable improvement in the terms of computational complexity.