Malicious content filtering based on semantic features

Kim, Semin; Min, Hyun‐Seok; Jeon, Jae Heung; Ro, Yong Man; Han, Seung-Wan

doi:10.1145/1655925.1656071

Cited by 8 publications

(7 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Finally, body part based detectors first extract previously defined semantic features that describe pornographic content (e.g. breast, belly, bottom) and then use in information as a set of vectors for further classification [24]. The main problem with this approach is its ambiguity and high rate of false positive detections due to small patch support and large appearance variations in the training set.…”

Section: A Detection Of Explicit Imagesmentioning

confidence: 99%

Keeping Children Safe Online With Limited Resources: Analyzing What is Seen and Heard

et al. 2021

View full text Add to dashboard Cite

It is every parent's wish to protect their children from online pornography, cyber bullying and cyber predators. Several existing approaches analyze a limited amount of information stemming from the interactions of the child with the corresponding online party. Some restrict access to websites based on a blacklist of known forbidden URLs, others attempt to parse and analyze the exchanged multimedia content between the two parties. However, new URLs can be used to circumvent a blacklist, and images, video, and text can individually appear to be safe, but need to be judged jointly. We propose a highly modular framework of analyzing content in its final form at the user interface, or Human Computer Interaction (HCI) layer, as it appears before the child: on the screen and through the speakers. Our approach is to produce Children's Agents for Secure and Privacy Enhanced Reaction (CASPER), which analyzes screen captures and audio signals in real time in order to make a decision based on all of the information at its disposal, with limited hardware capabilities. We employ a collection of deep learning techniques for image, audio and text processing in order to categorize visual content as pornographic or neutral, and textual content as cyberbullying or neutral. We additionally contribute a custom dataset that offers a wide spectrum of objectionable content for evaluation and training purposes. CASPER demonstrates an average accuracy of 88% and an F1 score of 0.85 when classifying text, and an accuracy of 95% when classifying pornography.INDEX TERMS Cyber-bullying, Cyber-grooming, Online Safety, Pornography filter, Real time agent.

show abstract

Section: A Detection Of Explicit Imagesmentioning

confidence: 99%

Keeping Children Safe Online With Limited Resources: Analyzing What is Seen and Heard

et al. 2021

View full text Add to dashboard Cite

show abstract

“…Unlike traditional region-based approaches, our approach does not rely on ROI detection, and directly identify pornographic images by evaluating our model on the test image at different positions and scales. Furthermore, while similar in spirit with the body partbased approaches [21], [26], our generic recognition model aims at more general and discriminative pornographic contents in images, which typically combine useful context with the key pornographic contents and thus can be more reliably recognized than the individual private body parts.…”

Section: Relative Work a Pornographic Image Recognitionmentioning

confidence: 99%

“…In this paper, we propose to train a generic region-based recognition model, which does not specialize in certain body part (e.g., breast) but aims at more general and discriminative visual patterns, for example, the female's upper body with exposed breast, the lower body with exposed private part, the close-up of sexual behaviour, etc. In contrast to the semantic feature approach [21], our target regions augment private body parts or sexual acts with useful context, and thus can be more reliably recognized. Furthermore, these regions are not required to contain complete private body part or sexual act, and some non-typical pornographic regions (e.g., a region containing only a part of breast) will also do.…”

Section: Introductionmentioning

confidence: 99%

“…Standard supervised learning requires bounding box annotations for the pornographic regions of interest. However, unlike the well-defined semantic features [21], the annotation process of our target regions is easily influenced by subjectivity, as different people may have dif-ferent understanding of useful context. Furthermore, different regions may have different degree of pornography -intuitively, a region containing a complete breast is more pornographic than the one containing a part of breast.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Pornographic Image Recognition via Weighted Multiple Instance Learning

Jin¹,

Yu-hui

Tan

2019

IEEE Trans. Cybern.

View full text Add to dashboard Cite

In the era of Internet, recognizing pornographic images is of great significance for protecting children's physical and mental health. However, this task is very challenging as the key pornographic contents (e.g., breast and private part) in an image often lie in local regions of small size. In this paper, we model each image as a bag of regions, and follow a multiple instance learning (MIL) approach to train a generic region-based recognition model. Specifically, we take into account the regions' degree of pornography, and make three main contributions. First, we show that based on very few annotations of the key pornographic contents in a training image, we can generate a bag of properly sized regions, among which the potential positive regions usually contain useful contexts that can aid recognition. Second, we present a simple quantitative measure of a region's degree of pornography, which can be used to weigh the importance of different regions in a positive image. Third, we formulate the recognition task as a weighted MIL problem under the convolutional neural network framework, with a bag probability function introduced to combine the importance of different regions. Experiments on our newly collected large scale dataset demonstrate the effectiveness of the proposed method, achieving an accuracy with 97.52% true positive rate at 1% false positive rate, tested on 100K pornographic images and 100K normal images.

show abstract

“…Hence, we believe that semantic concepts (detected from adult image) could be effective with high discriminatory power for the purpose of classifying naked and bikinied images. However, there are few attempts to make use of semantic feature and to analyze the effect of semantic feature compared with low-level visual feature for naked image classification [4]. In this paper, we present useful semantic features suitable for naked image classification.…”

Section: Introductionmentioning

confidence: 99%

Semantic Detection of Adult Image Using Semantic Features

Jeon

Choi

Min

et al. 2010

2010 4th International Conference on Multimedia and Ubiquitous Engineering

View full text Add to dashboard Cite

Recently, in the fields of internet and social networking, the classification and filtering of naked images has been receiving a significant amount of attention. In this paper, we propose a novel naked image classification which can make effective use of semantic features of a naked image. In addition, a novel measurement, termed accumulated distance ratio (ADR), is proposed in order to systematically analyze the effect of semantic features on improving classification performance, compared to the approach relying on low-level visual features. Extensive experiments have been carried out to assess the effectiveness of semantic features in naked image classification with realistic and challenging data set. The experimental result of the proposed approach using semantic features, for challenging data set, shows improvement up to 14% than the approach using low-level visual feature. Further, the proposed ADR measure has proven to be useful measure for analyzing the effect of semantic features for naked image classification.

show abstract

Malicious content filtering based on semantic features

Cited by 8 publications

References 11 publications

Keeping Children Safe Online With Limited Resources: Analyzing What is Seen and Heard

Keeping Children Safe Online With Limited Resources: Analyzing What is Seen and Heard

Pornographic Image Recognition via Weighted Multiple Instance Learning

Semantic Detection of Adult Image Using Semantic Features

Contact Info

Product

Resources

About