Classifying Objectionable Websites Based on Image Content

James Ze Wang, Jia Li, Gio Wiederhold, Oscar Firschein
Stanford University, Stanford, CA 94305
Abstract:

This paper describes IBCOW (Image-based Classification of Objectionable Websites), a system capable of classifying a website as objectionable or benign based on image content. The system uses WIPE (Wavelet Image Pornography Elimination) and statistics to provide robust classification of on-line objectionable World Wide Web sites. Semantically-meaningful feature vector matching is carried out so that comparisons between a given on-line image and images marked as "objectionable" and "benign" in a training set can be performed efficiently and effectively in the WIPE module. If more than a certain number of images sampled from a site is found to be objectionable, then the site is considered to be objectionable. The statistical analysis for determining the size of the image sample and the threshold number of objectionable images is given in this paper. The system is practical for real-world applications, classifying a Web site at a speed of less than 2 minutes each, including the time to compute the feature vector for the images downloaded from the site, on a Pentium Pro PC. Besides its exceptional speed, it has demonstrated higher than 97% sensitivity and 97% specificity in classifying a Web site based solely on images. Both the sensitivity and the specificity in real-world applications is expected to be higher because our performance evaluation is relatively conservative and surrounding text can be used to assist the classification process.

Full Paper in Color
(PDF, 1.2MB)

Full Paper in Color
(PostScript, 1.2MB)

Citation: James Z. Wang, Jia Li, Gio Wiederhold and Oscar Firschein, ``Classifying Objectionable Websites Based on Image Content,'' Lecture Notes in Computer Science, Special issue on interactive distributed multimedia systems and telecommunication services, Oslo, Norway, Thomas Plagemann and Vera Goebel (eds.), vol. 1483, pp. 113-124, Springer-Verlag, September 1998.

Copyright 1998 Springer-Verlag. Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work in other works, must be obtained from the Springer-Verlag.

Last Modified: 01-Mar-98 22:10:45 PST
1998, James Z. Wang