光学 精密工程, 2018, 26 (11): 2785, 网络出版: 2019-01-10
基于相机溯源的潜在不良视频通话预警
Early warning of illegal video chats based on camera source identification
相机溯源 光响应非均匀特性 集成学习 实时视频流 camera source identification photo response non-uniformity ensemble learning live video streaming
摘要
目前主流的不良视频检测大多基于视频内容分析, 属于计算密集型任务, 不利于巨量视频通话流的实时检测。为此, 本文试图从新的视角来阻断潜在不良视频的传播, 即实时视频流相机溯源。和传统检测内容的方法不同, 本文试图通过检测产生不良视频的相机源来阻断潜在不良视频的传播, 即一旦发现某个正在视频通话的手机产生过不良视频, 就发出安全警告进行阻断。该思路的基本假设是拍摄过不良视频的手机拥有者有更大概率利用同一部手机拍摄不良视频。该思路的核心问题是寻找一种实时、可靠的相机溯源方法。为此, 本文主要聚焦以下三方面工作: 建立了一个包含100部视频的数据库用于算法评估, 数据库中的视频来自25部不同型号、不同品牌的相机, 每一部视频注明了相机来源; 建立了一种简单、有效的视频相机溯源机制, 实现相机的在线实时溯源; 提出了一种多相机指纹特征集成决策模型, 实现可靠的相机溯源。实验结果显示, 所提相机溯源机制能满足相机溯源的实时性要求, 并且所提多相机指纹特征集成决策模型显著优于现有的单一模型, 对于安卓手机, 其视频相机溯源准确率达到98.161%, 验证了该思路的可行性。
Abstract
Majority of the techniques currently employed for detecting illegal videos were primarily based on video content analysis, which in turn was computationally intensive and cannot be applied for real-time detection of large video traffic flow. To address the said issue, the novel approach of real-time video source identification was proposed in this paper for preventing the transmission of illegal videos. In contrast to conventional approached based on content analysis, the proposed approach blocked the transmission of potentially illegal videos by detecting the camera sources that had previously produced illegal videos. If a video call was performed from a cellphone that had previously produced illegal videos, a security warning was issued for blocking the video transmission. The proposed approach was based on the assumption that video streams were more likely to be illegal if they have been produced from a camera that was used earlier for distributing illegal videos. Hence, the primary objective of the proposed approach was to develop a real-time and reliable method for camera source identification. To achieve this two objective, the following three aspects were taken into consideration. Firstly, a database comprising hundred videos was established for evaluation. The videos in the database were obtained from twenty-five cellphones of various brands and models, wherein each video indicated the corresponding camera source. Secondly, a simplified and feasible method was developed for real-time video source identification. Finally, a novel feature decision-making model with multiple integrated fingerprint features was incorporated for enhancing the reliability of video source identification. Obtained experimental results indicate that the proposed approach for camera source identification can meet real-time requirements and is significantly superior to conventional approaches. Furthermore, in the case of Android cellphones, the accuracy of camera source identification is found to be 98.161%, which corroborates the feasibility of the proposed approach.
马晓晨, 韦世奎, 蒋翔, 李晓飞. 基于相机溯源的潜在不良视频通话预警[J]. 光学 精密工程, 2018, 26(11): 2785. MA Xiao-chen, WEI Shi-kui, JIANG Xiang, LI Xiao-fei. Early warning of illegal video chats based on camera source identification[J]. Optics and Precision Engineering, 2018, 26(11): 2785.