Super-Resolution X-Ray Imaging Method and Apparatus
In one embodiment, a computing system may obtain a high-resolution X-ray image and a number of low-resolution X-ray images of an object of interest. The system may divide each of the low-resolution X-ray images into a number of low-resolution patches. Each low-resolution patch may be associated with a portion of the object of interest. The system may input a set of low-resolution patches associated with a same portion of the object of interest into a machine-learning model. Each low-resolution patch of the set may be from a different low-resolution X-ray image. The machine-learning model may output a high-resolution patch for the same portion of the object of interest. The system may compare the high-resolution patch outputted by the machine-learning model to a corresponding portion of the high-resolution X-ray image of the object of interest and adjust one or more parameters of the machine-learning model based on the comparison.
1 . A method comprising, by a computing system:
obtaining a high-resolution X-ray image and a plurality of low-resolution X-ray images of an object of interest;
dividing each low-resolution X-ray image of the plurality of low-resolution X-ray images into a plurality of low-resolution patches, wherein each low-resolution patch is associated with a portion of the object of interest;
inputting a set of low-resolution patches associated with a same portion of the object of interest into a machine-learning model, wherein each low-resolution patch of the set is from a different low-resolution X-ray image, and wherein the machine-learning model outputs a high-resolution patch for the same portion of the object of interest;
comparing the high-resolution patch outputted by the machine-learning model to a corresponding portion of the high-resolution X-ray image of the object of interest; and
adjusting one or more parameters of the machine-learning model based on the comparison.
2 . The method of claim 1 , further comprising, prior to obtaining the high-resolution X-ray image and the plurality of low-resolution X-ray images:
determining a target resolution for super-resolution X-ray images outputted by the machine-learning model, wherein the target resolution of the super-resolution X-ray images has a ratio to a resolution of the high-resolution X-ray images, and wherein the resolution of the high-resolution X-ray images has the same ratio to a resolution of the low-resolution X-ray images.
3 . The method of claim 1 , wherein the machine-learning model is a regression classifier, an artificial neural network, or a support vector machine.
4 . The method of claim 1 , wherein the high-resolution X-ray image has a resolution equal to an X-ray sensor resolution.
5 . The method of claim 1 , wherein each low-resolution X-ray image of the plurality of low-resolution X-ray images has a resolution lower than an X-ray sensor resolution.
6 . The method of claim 1 , wherein the plurality of low-resolution images capture the object of interest from at least two perspectives.
7 . The method of claim 1 , wherein the plurality of low-resolution images comprises at least two images obtained by:
changing an imaging condition;
acquiring a low-resolution X-ray image using the imaging condition; and
repeating the changing and acquiring steps until all images of the plurality of low-resolution X-ray images are obtained.
8 . The method of claim 7 , wherein changing the image condition results in a sub-pixel shift from one low-resolution X-ray image to a next low-resolution X-ray image of the plurality of X-ray images.
9 . The method of claim 7 , wherein the imaging condition comprises an incident angle of an X-ray beam emitted from an X-ray source of an X-ray system.
10 . The method of claim 9 , wherein the incident angle is changed by varying a position of the X-ray source.
11 . The method of claim 7 , wherein the imaging condition comprises a position of an X-ray detector.
12 . The method of claim 7 , wherein the imaging condition comprises a position of the object of interest.
13 . The method of claim 1 , further comprising:
obtaining a set of high-resolution X-ray images of a second object of interest;
dividing each of the high-resolution x-ray images in the set into high-resolution patches;
for each patch region, inputting one or more of the high-resolution patches for the patch region, to an instance of the machine-learning model to generate a super-resolution patch for the patch region; and
generating a super-resolution X-ray image of the second object of interest based on the super-resolution patches for the patch regions.
14 . The method of claim 13 , wherein the super-resolution patch for the patch region is determined further based on one or more of the high-resolution patches for patch regions that are nearest-neighbors to the patch region.
15 . The method of claim 13 , further comprising:
displaying the super-resolution X-ray image on a monitor.
16 . The method of claim 13 , further comprising:
detecting one or more defects in the object of interest based on the super-resolution X-ray image.
17 . The method of claim 13 , further comprising:
monitoring a manufacturing process of the object of interest based on the super-resolution X-ray image.
18 . The method of claim 13 , wherein the super-resolution X-ray image has a higher resolution than the high-resolution X-ray image by at least a factor of two.
19 . One or more computer-readable non-transitory storage media embodying software that is operable when executed to:
obtain a high-resolution X-ray image and a plurality of low-resolution X-ray images of an object of interest;
divide each low-resolution X-ray image of the plurality of low-resolution X-ray images into a plurality of low-resolution patches, wherein each low-resolution patch is associated with a portion of the object of interest;
input a set of low-resolution patches associated with a same portion of the object of interest into a machine-learning model, wherein each low-resolution patch of the set is from a different low-resolution X-ray image, and wherein the machine-learning model outputs a high-resolution patch for the same portion of the object of interest;
compare the high-resolution patch outputted by the machine-learning model to a corresponding portion of the high-resolution X-ray image of the object of interest; and
adjust one or more parameters of the machine-learning model based on the comparison.
20 . A system comprising:
one or more processors; and
one or more computer-readable non-transitory storage media coupled to one or more of the processors and comprising instructions operable when executed by one or more of the processors to cause the system to:
obtain a high-resolution X-ray image and a plurality of low-resolution X-ray images of an object of interest;
divide each low-resolution X-ray image of the plurality of low-resolution X-ray images into a plurality of low-resolution patches, wherein each low-resolution patch is associated with a portion of the object of interest;
input a set of low-resolution patches associated with a same portion of the object of interest into a machine-learning model, wherein each low-resolution patch of the set is from a different low-resolution X-ray image, and wherein the machine-learning model outputs a high-resolution patch for the same portion of the object of interest;
compare the high-resolution patch outputted by the machine-learning model to a corresponding portion of the high-resolution X-ray image of the object of interest; and
adjust one or more parameters of the machine-learning model based on the comparison.