Adult Image Detection Combining BoVW Based on Region of Interest and Color Moments

Liu Yizhi (1, 2, 3) , Lin Shouxun (1) , Tang Sheng (1) , Zhang Yongdong (1)

1 Laboratory of Advanced Computing Research, Institute of Computing Technology
2 GSCAS - Graduate School of the Chinese Academy of Sciences
3 Institute of Computer Science and Engineering

To prevent pornography from spreading on the Internet effectively, we propose a novel method of adult image detection which combines bag-of-visual-words (BoVW) based on region of interest (ROI) and color moments (CM). The goal of BoVW is to automatically mine the local patterns of adult contents, called visual words. The usual BoVW method clusters visual words from the patches in the whole image and adopts the weighting schemes of hard assignment. However, there are many background noises in the whole image and soft-weighting scheme is better than hard assignment. Therefore, we propose the method of BoVW based on ROI, which includes two perspectives. Firstly, we propose to create visual words in ROI for adult image detection. The representative power of visual words can be improved because the patches in ROI are more indicative to adult contents than those in the whole image. Secondly, soft-weighting scheme is adopted to detect adult images. Moreover, CM is selected by evaluating some commonly-used global features to be combined with BoVW based on ROI. The experiments and the comparison with the state-of-the-art methods show that our method is able to remarkably improve the performance of adult image detection.

Mots clés: Adult image detection; bag-of-visual-words (BoVW); region of interest (ROI); soft-weighting; color moments

Domaines: Bibliothèque électronique [cs.DL]

Liu Yizhi, Lin Shouxun, Tang Sheng, Zhang Yongdong. Adult Image Detection Combining BoVW Based on Region of Interest and Color Moments. 6th IFIP TC 12 International Conference on Intelligent Information Processing (IIP), Oct 2010, Manchester, United Kingdom. pp.316-325, ⟨10.1007/978-3-642-16327-2_38⟩. ⟨hal-01055069⟩