Image ciphertexts classification method based on ghost imaging and intraclass-interclass difference

In this paper, based on ghost imaging encryption, the preservation of Manhattan distance feature in ciphertext compared with plaintext is analyzed by utilizing the intraclass-interclass difference of image classification, and a classification method for image ciphertexts is proposed. After calculating Manhattan distance for both plaintexts and ciphertexts, respectively, the intraclass-interclass difference can be determined. The image that minimizes the intraclass-interclass difference is taken as the centroid to verify the consistency of the classification for various plaintext-ciphertext pairs under the same operation. The feasibility of proposed method is verified by numerical simulations, that the values of ACC and Weighted-F2 can be up to 90% when the MNIST is adopted as the test dataset. The whole process can be regarded as a kind of classification process by homomorphic encryption, however, different from the traditional homomorphic encryption methods based on mathematical model, the proposed method is accomplished based on the optical theory, and it does not require a lot of pre-training through models such as deep learning and neural networks, that means, reducing the computational expenses.