Anonymizing personally identifiable information in image data
A computer includes a processor and a memory. The memory stores instructions executable by the processor. The computer identifies personally identifiable information (PII) in an initial image frame. The computer encodes a region containing the PII of the initial image frame to a latent vector with a lower dimensionality than the region. The computer anonymizes the region from the initial image frame, resulting in a modified image frame. The computer associates: the latent vector with the modified image frame.
1 . A computer comprising a processor and a memory, the memory storing instructions executable by the processor to:
identify personally identifiable information (PII) in an initial image frame;
encode a region containing the PII of the initial image frame to a latent vector with a lower dimensionality than the region;
anonymize the region from the initial image frame, resulting in a modified image frame;
associate the latent vector with the modified image frame;
execute a machine-learning perception algorithm using the modified image frame with the latent vector as an input; and
actuate a component of a vehicle based on the modified image frame with the latent vector and based on an output of the machine-learning perception algorithm.
2 . The computer of claim 1 , wherein the instructions further include instructions to receive the initial image frame from a camera of the vehicle.
3 . The computer of claim 1 , wherein the instructions further include instructions to execute the machine-learning perception algorithm using the initial image frame and the modified image frame with the latent vector as inputs.
4 . The computer of claim 1 , wherein the instructions further include instructions to embed the latent vector in the modified image frame.
5 . The computer of claim 1 , wherein the modified image frame with the latent vector has a same dimensionality as the initial image frame.
6 . The computer of claim 1 , wherein the instructions further include instructions to transmit the modified image frame with the latent vector to a remote server.
7 . The computer of claim 6 , wherein the instructions further include instructions to refrain from transmitting the initial image frame to the remote server.
8 . The computer of claim 1 , wherein the PII includes a face of a person.
9 . The computer of claim 8 , wherein the latent vector encodes non-personally identifying data about the face.
10 . The computer of claim 1 , wherein the PII includes a depiction of text.
11 . The computer of claim 1 , wherein the instructions further include instructions to execute a machine-learning encoder on the PII to encode the PII to the latent vector.
12 . The computer of claim 11 , wherein the machine-learning encoder is trained as part of an encoder-decoder architecture to reconstruct the region of the initial image frame.
13 . The computer of claim 12 , wherein the encoder-decoder architecture is trained using a loss function that includes a reconstruction loss measuring similarity between an output of the encoder-decoder architecture and the region of the initial image frame inputted to the encoder-decoder architecture.
14 . The computer of claim 11 , wherein the machine-learning perception algorithm is jointly trained with the machine-learning encoder.
15 . The computer of claim 14 , wherein the machine-learning perception algorithm and the machine-learning encoder are jointly trained with the latent vector outputted by the machine-learning encoder being inputted to the machine-learning perception algorithm.
16 . The computer of claim 14 , wherein the machine-learning perception algorithm and the machine-learning encoder are jointly trained using a loss function that includes a loss measuring an accuracy at correctly identifying objects in input image frames.
17 . A method comprising:
identifying personally identifiable information (PII) in an initial image frame;
encoding a region containing the PII of the initial image frame to a latent vector with a lower dimensionality than the region;
anonymizing the region from the initial image frame, resulting in a modified image frame;
associating the latent vector with the modified image frame;
executing a machine-learning perception algorithm using the modified image frame with the latent vector as an input; and
actuating a component of a vehicle based on the modified image frame with the latent vector and based on an output of the machine-learning perception algorithm.
18 . A computer comprising a processor and a memory, the memory storing instructions executable by the processor to:
identify personally identifiable information (PII) in an initial image frame;
encode a region containing the PII of the initial image frame to a latent vector with a lower dimensionality than the region;
anonymize the region from the initial image frame, resulting in a modified image frame;
associate the latent vector with the modified image frame; and
execute a machine-learning perception algorithm using the initial image frame and the modified image frame with the latent vector as inputs.
19 . A computer comprising a processor and a memory, the memory storing instructions executable by the processor to:
identify personally identifiable information (PII) in an initial image frame;
encode a region containing the PII of the initial image frame to a latent vector with a lower dimensionality than the region;
anonymize the region from the initial image frame, resulting in a modified image frame;
associate the latent vector with the modified image frame;
execute a machine-learning encoder on the PII to encode the PII to the latent vector; and
execute a machine-learning perception algorithm using the modified image frame with the latent vector as an input, wherein the machine-learning perception algorithm is jointly trained with the machine-learning encoder.