Machine learning-based digital holography device and method for operating same
A machine learning-based digital holography device and a method for operating same are disclosed. The digital holography method according to one embodiment comprises the steps of: obtaining an optical image including two-dimensional information about a sample; and on the basis of the optical image, generating a holographic image including three-dimensional information about the sample.
1. A digital holographic method including:
obtaining a bright-field (BF) image comprising two-dimensional (2D) information of a sample; and
generating a hologram image comprising three-dimensional (3D) information of the sample by inputting the BF image to a neural network, the neural network being trained by statistically learning a relationship between a training BF image according to a depth of a training sample and a corresponding training hologram image of the training sample.
2. The digital holographic method of claim 1 , wherein the BF image is captured by irradiating the sample with white light.
3. The digital holographic method of claim 1 , wherein
the training BF image and the corresponding training hologram image are obtained simultaneously by illuminating the same training sample with white light and monochromatic beam at the same time.
4. The digital holographic method of claim 1 , further comprising:
obtaining 3D information of the sample by numerically reconstructing the hologram image in a depth direction.
5. The digital holographic method of claim 1 , further comprising:
obtaining a light scattering pattern by accumulating light intensity distributions of the hologram image in the depth direction; and
extracting at least one of a real focal length or a virtual focal length of the sample, based on the light scattering pattern.
6. A digital holographic apparatus comprising:
a memory containing instructions; and
a processor configured to execute the instructions,
wherein, when the instructions are executed by the processor, the processor is configured to:
generate a hologram image comprising three-dimensional (3D) information of a sample by inputting bright-field (BF) image comprising two-dimensional (2D) information of the sample to a neural network, the neural network being trained by statistically learning a relationship between a training BF image according to a depth of a training sample and a corresponding training hologram image of the training sample.
7. The digital holographic apparatus of claim 6 , wherein the BF image is captured by irradiating the sample with white light.
8. The digital holographic apparatus of claim 6 , wherein
the training BF image and the corresponding training hologram image are obtained simultaneously by illuminating the same training sample with white light and monochromatic light at the same time.
9. The digital holographic apparatus of claim 6 , wherein the processor is configured to:
obtain 3D information of the sample by numerically reconstructing the hologram image in the depth direction.
10. The digital holographic apparatus of claim 6 , wherein the processor is configured to:
obtain a light scattering pattern by accumulating light intensity distributions of the hologram image in the depth direction; and
extract at least one of a real focal length or a virtual focal length of the sample based on the light scattering pattern.
11. The digital holographic method of claim 1 , wherein the relationship includes a relationship between a degree of defocusing in the training BF image according to the depth and an interference fringe of the training hologram image.
12. The digital holographic apparatus of claim 6 , wherein the relationship includes a relationship between a degree of defocusing in the training BF image according to the depth and an interference fringe of the training hologram image.