Granted Patent B2

US 12,412,277 · App. 18/364,898 · Granted Sep 9, 2025

Inventors: Shinwook Kim (Daejeon, KR); Youngchun Youk (Daejeon, KR)

Assignee: KOREA AEROSPACE RESEARCH INSTITUTE

G06T7/13G06T7/168G06T2207/20056G06T2207/20081

Abstract

Claims (34)

1. A method of measuring spatial resolution of electro-optical system, the method comprising:

obtaining raw edge spread function (ESF) data from an edge image obtained by using the electro-optical system;

obtaining a corrected ESF curve by inputting the raw ESF data to a deep learning model;

obtaining a line spread function (LSF) curve by differentiating the corrected ESF curve;

obtaining a modulation transfer function (MTF) curve by Fourier-transforming the LSF curve; and

obtaining a MTF value from the MTF curve.

2. The method of claim 1 , further comprising

obtaining a plurality of sample ESF data;

obtaining a plurality of sample MTF values respectively corresponding to the plurality of sample ESF data;

selecting some sample ESF data from among the plurality of sample ESF data based on the plurality of sample MTF values;

obtaining a ground-truth ESF curve by averaging the some sample ESF data; and

training the deep learning model by using at least one of the plurality of sample ESF data and the ground-truth ESF curve as training data.

3. The method of claim 2 , wherein the obtaining of a plurality of sample ESF data comprises:

obtaining a plurality of sample images by capturing images of a plurality of target samples, the plurality of target samples each having a knife-edge located at a preset pixel distance; and

obtaining each of the plurality of sample ESF data from the plurality of sample images.

4. The method of claim 3 , wherein the obtaining of the ground-truth ESF curve comprises:

arranging the some sample ESF data based on the preset pixel distance; and

obtaining the ground-truth ESF curve by averaging the arranged some sample ESF data.

5. The method of claim 2 , wherein the deep learning model is trained based on a central sample ESF data obtained from a target sample in which a knife-edge is located at a center among the plurality of sample ESF data, and the ground-truth ESF curve.

6. The method of claim 2 , wherein the selecting of the some sample ESF data comprises:

selecting a plurality of intermediate MTF values located in a middle when the plurality of sample MTF values are arranged in order of size; and

selecting the some sample ESF data respectively corresponding to the plurality of intermediate MTF values from among the plurality of sample ESF data.

7. The method of claim 2 , wherein the training of the deep learning model comprises:

obtaining an output ESF curve by inputting one of the at least one of the plurality of sample ESF data to the deep learning model; and

training the deep learning model to reduce a difference between the output ESF curve and the ground-truth ESF curve.

8. The method of claim 2 , further comprising:

obtaining a ground-truth LSF curve by differentiating the ground-truth ESF curve; and

obtaining a ground-truth MTF curve by Fourier-transforming the ground-truth LSF curve.

9. The method of claim 1 , wherein the MTF value is a value corresponding to a Nyquist frequency value in the MTF curve.

10. A non-transitory computer-readable medium storing a computer program including instructions that, when executed by a processor, causes a computer to execute the method of claim 1 .

11. An apparatus for measuring spatial resolution of electro-optical system, the apparatus comprising:

a memory; and

a processor communicatively connected to the memory, and configured to obtain raw edge spread function (ESF) data from an edge image obtained by using the electro-optical system, obtain a corrected ESF curve by inputting the raw ESF data to a deep learning model, obtain a line spread function (LSF) curve by differentiating the corrected ESF curve, obtain a modulation transfer function (MTF) curve by Fourier-transforming the LSF curve, and obtain an MTF value from the MTF curve.

12. The apparatus of claim 11 , wherein the processor is further configured to obtain a plurality of sample ESF data, obtain a plurality of sample MTF values respectively corresponding to the plurality of sample ESF data, select some sample ESF data from among the plurality of sample ESF data based on the plurality of sample MTF values, obtain a ground-truth ESF curve by averaging the some sample ESF data, and train the deep learning model by using at least one of the plurality of sample ESF data and the ground-truth ESF curve, as training data.

Assignments (1)

Priority Claims (1)

Continuity (1)

References Cited (10)