量化SIFT和同态加密的隐私保护图像检索方法

Privacy preserving image retrieval method based on binary SIFT and homomorphic encryption

提出了一种隐私语义保持的图像内容检索方法,将加密图像中隐私保持尺度不变特征变换(SIFT)的提取方法和二进制SIFT算法融合在一起,不仅保证了上传到服务器端的图像是加密的,同时又能在加密空间保持其隐私语义。对图像进行Paillier同态加密,保证了图像在服务器端和传输过程中的安全性,在加密域提取SIFT特征,并将其用二进制表示,减少存储空间和计算复杂度。实验证明:经原始图像特征提取后生成的二进制SIFT在稳健性测试中获得良好的效果,并且与加密图像特征提取后生成的二进制SIFT保持等距,在明文域和密文域中保持了图像搜索匹配的准确性,在匹配效率上得到提高。

A privacy preserving image content retrieval method,which combines privacy preserving SIFT extraction method in encrypted image with binary SIFT algorithm is proposed.In this way,it can not only guarantee the image uploaded to the server is encrypted,but also keep its semantic privacy in encrypted domain.Image is encrypted by using Paillier homomorphic encryption,this can ensure security of image on server end.And SIFT features are extracted in the encrypted domain,after that the features are converted to a binary representation,reduce storage space and computational complexity.Experimental results show that the binary SIFT extracted from encrypted domain performs well in robustness evaluation,and the similarity distances between binary SIFT generated from the original image and the binary SIFT generated from the encrypted image keep the same,and ensure accuracy of image searching and matching in plaintext and ciphertext domain,and the matching efficiency is improved.