A novel violation detection method of live video using fuzzy support vector machine
- PDF / 1,161,125 Bytes
- 10 Pages / 595.276 x 790.866 pts Page_size
- 92 Downloads / 178 Views
ORIGINAL RESEARCH
A novel violation detection method of live video using fuzzy support vector machine Chao Yuan1,2 · Jie Zhang1,3 Received: 14 July 2020 / Accepted: 10 October 2020 © Springer-Verlag GmbH Germany, part of Springer Nature 2020
Abstract In order to prevent the illegal videos from being posted on the Internet and causing adverse effects, video sites need to manually review each newly released video. The manual review is time-consuming and labor-intensive, and is prone to omissions. Against this background, this article intends to propose a method for automatically detecting illegal content in videos. Automatic video detection can greatly reduce the work of auditors and improve detection efficiency. This study proposes a multi-modal fusion feature violation video detection method using fuzzy support vector machine (FSVM). First, extract multiple modal features of live video, including still image features, motion features, and audio features. Secondly, FSVM is used to classify the feature data of various modalities to obtain the classification results under different modalities. Finally, the classification results in different modes are merged to obtain the final decision result. The innovation of this study is that the introduction of multiple modal features enriches the sample information, making the sample information more comprehensive. Which is easy to distinguish. The classifier FSVM is based on the traditional SVM to assign a degree of membership to each sample, thereby reducing the impact of isolated points and noise on the optimal decision surface. Experiments show that this study improves the detection efficiency of illegal videos and can meet the requirements of practical applications. Keywords Violation video detection · Multi-modal features · FSVM · Noise immunity
1 Introduction With the development of the live broadcasting industry, the security problems of live broadcasting are gradually emerging. At present, the supervision of each live broadcast platform is basically in the state of manpower supervision. Manpower supervision is mainly carried out through user reports and platform supervisors. However, the number of live broadcast rooms in the same period is huge. Relying solely on human supervision can no longer meet the regulatory requirements for live broadcasting. At present, there are many network firewall software used to prevent minors * Jie Zhang [email protected] 1
College of Economics and Management, Nanjing University of Aeronautics and Astronautics, 29 Jiangjun Avenue, Nanjing 211100, People’s Republic of China
2
School of Design, Jiangnan University, 1800 Lihu Avenue, Wuxi 214122, Jiangsu, People’s Republic of China
3
Nanjing University of the Arts, No. 74 Beijing West Road, Nanjing 210013, Jiangsu, People’s Republic of China
from browsing bad websites and web pages. Most of these software use pornographic URLs and sensitive words. But neither of these methods can effectively detect and filter online porn videos. Starting from the illegal video itself,
Data Loading...
