Privacy protection image classification method based on domain adaption

Abstract

The invention belongs to the technical field of information security, and provides a privacy protection image classification method based on domain adaption. In the first part, firstly, data is preprocessed so as to standardize the data, and a data set requires at least one labelled data set and a plurality of label-free data sets; then a differential privacy definition suitable for the image data is given, and noise disturbance is added to the data to meet the requirement of differential privacy; and finally, availability measurement is performed on the data subjected to noise disturbance so as to ensure the availability of the data. In a second part, firstly, a proper feature transformation dimension is determined by using a moment generating function of probability distribution, and data is mapped to a new feature space; secondly, an image classification model Cy is trained in the mapped feature space by using the labeled data obtained in the first part; then, the indistinguishability of a data subordinate data set is enhanced by using a generative adversarial network; and finally, the label-free data is classified by using Cy.

Description

technical field [0001] The invention relates to a privacy protection image classification method based on a domain self-adaptive method, which belongs to the technical field of information security. Background technique [0002] The great success achieved in the field of machine learning in the past decade relies on the improvement of computing power, the emergence of new machine learning paradigms (deep neural networks) and the support of large-scale data. Based on this, researchers continue to propose new algorithms and models and apply them to various fields. Among them, the neural network model has made significant progress in the field of image recognition and classification. Under the same task, the accuracy of the neural network model exceeds traditional methods such as support vector machines, and also exceeds manual classification. Criminal suspects who have been at large for many years have been identified by image classification technology supported by machine le...

AI Technical Summary

Problems solved by technology

[0004] However, while research in the field of image classification is booming, some studies have found that machine learning models such as neural networks can "memorize" sensitive information in image data, and can infer and restore some sensitive information in image data by calculating the gradient information of the model. For example, image classification learning in the medical field may lead to the leakage of private information about patients' illnesses

Images