A Novel Skin Tone Detection Algorithm for Contraband Image Analysis

Abstract

This paper examines skin tone detection algorithms used by first responder forensic tools such as File Hound. File Hound is a "field analysis" software application that is currently being used by over 100 law enforcement agencies, both internationally and domestically. It is mainly used in forensic investigations to search and identify pornographic images from a hard drive. Since the conception of File Hound, several steps have been taken to improve its performance and expand its features. One such feature is a skin tone detection filter that can identify images with a large skin color count from the aggregate image results found by File Hound. This filter is based on the idea that there is a positive correlation between images with a large skin color count and images that are pornographic in nature. A novel skin tone detection filter was developed and this filter was tested against random images obtained from the Compaq Image database for skin tone detection. The results of the test are encouraging in terms of accuracy and low error rates: type I = 20.64%, type II = 0.81%, accuracy = 78.55%.