Image Security Attribute Inference Method Based on Class Attribute Portrait

Access control, the last line of defense for information security, ensures that legitimate users can access resources within only a certain scope of authority. The extraction of high-quality attributes is the key to effectively implementing access control. As a type of unstructured data, the content attribute of an image is highly abstract. The class label of objects in an image can be obtained using image classification based on deep learning. However, this method cannot provide any information about unlearned classes. To infer the known class similar to unknown-class images, this paper proposes an image-oriented class attribute portrait. For a certain known class, a support vector machine is used to learn a classifier in the semantic description space. The class name, description space, and classifier are described as the class attribute portrait, which can be used to infer the content attribute of unknown images. Accordingly, the access control policy is determined. Experimental results confirm the effectiveness of the proposed method for the content attribute inference of unknown-class images.