基于深度神经网络的图像理解安全研究现状

Research Status of Image Understanding Security Based on Deep Neural Network

近年来,随着深度神经网络DNN的应用在计算机图形学领域中达到了较高的识别性能,对于人工智能识别与人眼识别之间的差别逐渐成为议题,同时也引出了包括基于DNN的图像理解的反知觉性所造成的安全隐患等问题。这个问题的根本原因在于深度神经网络的表达性。该文从图像理解原理以及可能导致的安全漏洞开始,分析了这些漏洞可能导致的问题以及一些防范方案,也提出了对未来图像理解安全问题的前景展望。

In recent years,with the application of deep neural network DNN in the field of computer graphics to achieve high recognition performance,the difference between artificial intelligence recognition and human eye recognition has gradually become an issue,and it has also led to including DNN-based The security risks caused by the counter-perception of image understanding.The root cause of this problem lies in the expressiveness of deep neural networks.This article starts with the principles of image understanding and the security vulnerabilities that may result,analyzes the problems that these vulnerabilities may cause and some prevention schemes,and also puts forward the prospects for the future image understanding security issues.