Facial recognition faces privacy issues. Homomorphic encryption ensures secure data processing without exposure, improving both accuracy and efficiency.

Step 1: Generate the keys

Please generate a homomorphic encryption public-private key pair, and securely store the private key yourself.

Step 2: Send Public Key.

The user sends his public key to the server.

Step 3: Select register picture

After selecting the registration image, the server has already saved the encrypted feature values of the image.

select a picture:

image 1
image 2
image 3

Step 4: Select request picture

Select an image for comparison.

select a picture:

image 1
image 2
image 3

Step 5: Encrypt the 'request image' using a public key

The user uses their own public key to encrypt the image.

Step 6: Send the encrypted image data

The user sends their encrypted image to the service provider.

Step 7: Start encrypted computation, awaiting analysis results.

The user selects the method for encrypted computation, using the ResNet18 network for encrypted inference to extract feature vectors, and calculates the distance with the encrypted feature vector of the template.

  • Gewu Acceleration Card Computation: Perform encrypted computation using the FPGA acceleration card from the Gewu platform.
  • Plaintext Computation: Perform computation on the unencrypted image.

Step 8: Use the private key to decrypt the analysis results.

The user receives the computation results, uses their private key to decrypt, and obtains the plaintext diagnostic results.