Article Preview
TopIntroduction
Due to the increased development in technologies such as smart gadgets (including phones, watches, and television sets), access to internet resources (including pornographic contents) has been made simpler. Also, improvements in the storage technologies have led to the springing up of vast repositories of pornographic contents that are easily accessible online. Moreover, most of these repositories store contents that border on child sexual abuse, which is a significant issue for law enforcers. Access to pornographic content could lead to hampering of interpersonal, financial, and occupational functioning of the persons involved. Additionally, it could also lead to emotional problems and sexual dissatisfaction (Short, Black, Smith, Wetterneck, & Wells, 2012). As a result of these issues, there has been an increased interest in the development of an automated solution for detection of pornographic content. Such a solution can help in the detection of inappropriate behavior via surveillance cameras in a given environment. Additionally, the solution can be employed in the prevention of upload or access of pornographic contents for specific demographics (like children) or environments (like the corporate environment and the educational environment). Also, this can be seen with measures general social media networks are employing in preventing the upload of explicit contents (Moustafa, 2015). In the detection of child sexual abuse materials, law enforcers could use such a solution to identify pornographic contents on a storage media easily (Perez, et al., 2017). An issue that arises in the development of adult content detection systems is the subjective nature of the definition of what is pornographic. For the purpose of this study, the definition of pornography as ”any sexually explicit material with the aim of sexual arousal or fantasy” (Short, Black, Smith, Wetterneck, & Wells, 2012) shall be used. Given the full interest of society as a whole, there has been the development of different schemes to prevent access to pornographic materials. One such scheme is filtering of keywords from URLs. This scheme relies on the fact that most websites on the internet that provide explicit contents use words that describe these contents (such as porn and free sex) so, blocks access to them (Dinh, Ngo, & Vu, 2015). But this scheme results in a high false negative rate because smart explicit content may use normal words that may not invoke the thought that they provide explicit contents but may actually be providing such contents. Also, this scheme may result in medical websites being blocked since medical information relating to reproductive health can be described as explicit. Another scheme employed is filtering access to prohibited sites by collecting a list of adult website addresses. This scheme involves building an extensive database of known websites that provide explicit content so that anytime a website address is entered, that given address is matched against the entries in the database and access is denied if there is a match in the database (Cohen-Almagor, 2013). This scheme depends on the knowledge of popular explicit contents’ website address. As a result, if a popular website address is changed or a new website created with a different website address, this scheme will not be able to block access to the website address since it would not be found in its database. The third scheme involves a content-based analysis for detecting explicit content. This scheme involves checking the contents (images, video, text) of websites to detect explicit content. This scheme allows for a cogent analysis of the actual content irrespective of descriptions of the contents like the website address. A lot of research has been put into this area using different approaches in detecting explicit content (Yan, Liu, Xie, Liao, & Yin, 2014; Moustafa, 2015; Bhatti, et al., 2018; Perez-Meana, 2011). Explicit Content Detection (ECD) is the process of classifying pictures or videos containing and displaying genital elements of the human body from any picture or video. Methods that use the content analysis-based scheme are classified into two categories being the recently developed learning techniques and the traditional region of interest techniques. In this study, a critical look would be given to the various steps involved in using each of the above-mentioned techniques.