Computer Vision Method for Detecting Adult-Oriented Content in Images

As people started using the Internet a huge volume of data has become available on-line. Now it is impossible to efficiently control its content that may be unwanted or inappropriate for children. This topic is even more challenging when it comes to images and video sequences. Therefore, the problem of detecting an adult-oriented content in images still remains unsolved and is widely addressed by many researchers. In this paper we propose a low-level method for filtering such unwanted elements. It is based on a novel approach to skin segmentation that engages Gaussian distributions. The algorithm is self-calibrating and does not require any pre-defined thresholds and parameters, which is a certain disadvantage of many skin segmentation algorithms.