RGB-NIR dual camera face anti-spoofing method
A method of face anti-spoofing, comprising, receiving a near infra-red facial image, having a near infrared channel, receiving a red-green-blue facial image, having a red channel, a green channel and a blue channel, generating a synthetic three channel image based on the near infrared channel, the red channel, the green channel and the blue channel and training a deep neural network based on the synthetic three channel image.
1. A method of facial recognition, comprising:
receiving a near infrared facial image, having a near infrared channel;
receiving a red-green-blue facial image, having a red channel, a green channel and a blue channel;
generating a synthetic three channel image, based on the near infrared channel, the red channel, the green channel and the blue channel, having a first new channel, a second new channel, and a third new channel; and
processing a deep neural network based on the synthetic three channel image wherein the near infrared facial image and the red-green-blue facial image are fed into the deep neural network.
2. The method of facial recognition of claim 1 , wherein the first new channel comprises the near infrared channel, the second new channel comprises the near infrared channel and the third new channel comprises an average of the red channel, the green channel and the blue channel.
3. The method of facial recognition of claim 1 , wherein the first new channel comprises the near infrared channel, the second new channel comprises an average of the green channel and the blue channel and the third new channel comprises the red channel.
4. The method of facial recognition of claim 1 , further comprising:
converting the red-green-blue facial image to a luma-blue-red facial image having a luma channel, the blue channel and the red channel;
replacing the luma channel with the near infrared channel; and
converting the luma-blue-red facial image to a second red-green-blue facial image.
5. The method of facial recognition of claim 1 , further comprising:
converting the red-green-blue facial image to a luma-blue-red image having a luma channel, the blue channel and the red channel;
replacing the luma channel with an average of the near infrared channel and the luma channel; and
converting the luma-blue-red facial image to a second red-green-blue facial image.
6. The method of facial recognition of claim 1 , further comprising:
converting the red-green-blue facial image to a luma-blue-red facial image having a luma channel, the blue channel and the red channel;
decomposing the luma channel in a frequency domain;
decomposing the near infrared channel in the frequency domain;
combining high frequency components of the decomposed luma channel and high frequency components of the decomposed near infrared channel;
combining low frequency components of the decomposed luma channel and low frequency components of the decomposed near infrared channel; and
reconstructing a new synthetic image from the combined high frequency components and the combined low frequency components.
7. The method of facial recognition of claim 1 , wherein processing the deep neural network based on the synthetic three channel image wherein the near infrared facial image and the red-green-blue facial image are fed into the deep neural network comprises
fusing at an intermediate layer a first feature map based on the near infrared channel and a second feature map based on the red channel, the green channel and the blue channel to form a fused feature map,
wherein the fused feature map is fused at a convolutional layer at an intermediate level of the deep neural network.
8. The method of facial recognition of claim 7 , wherein the convolutional layer fusing is performed on a depth-wise and point-wise basis.
9. The method of facial recognition of claim 7 , wherein the first feature map and the second feature map are processed in parallel.