Mark detection method and computer program
An encoding step and an image selection step are performed by a computer 20 . In the encoding step, a plurality of partial images D 2 of an object are input to a learned VAE model M, and the plurality of partial images D 2 are encoded to a plurality of latent variables in a multi-dimensional latent space. In the image selection step, one or more partial images D 2 that are likely to include an alignment mark are selected on the basis of a distribution of the plurality of latent variables in the latent space. Thus, a partial image that is likely to include an alignment mark is automatically detected by the computer 20 with the use of the VAE model M. This can lighten a burden on an operator who searches for an alignment mark.
1 . A mark detection method, executed by a computer, for detecting an alignment mark from an object, the method comprising steps of:
A) inputting a plurality of partial images of the object to a learned VAE model and encoding the plurality of partial images to a plurality of latent variables in a multi-dimensional latent space; and
B) selecting one or more partial images that are likely to include the alignment mark based on a distribution of the plurality of latent variables in the latent space,
wherein the step B) includes steps of:
a) calculating a distance from each of the latent variables to a nearest latent variable in the latent space; and
b) selecting one or a predetermined number of partial images corresponding to one or a predetermined number of latent variables each having the distance larger than those of the other latent variables, as the one or more partial images that are likely to include the alignment mark.
2 . The mark detection method according to claim 1 , further comprising a step of
C) extracting a partial image including a characteristic shape from an image of the object, before the encoding step, and
wherein in the step A), the partial image extracted in the partial-image extraction step is input to the VAE model.
3 . The mark detection method according to claim 1 , further comprising a step of
D) removing a partial image including a probe mark, a flaw, or a particle from the plurality of partial images by using a classifier created by supervised learning.
4 . The mark detection method according to claim 1 , further comprising a step of a learning step of constructing or updating the VAE model by machine learning, and
wherein in the step E), an input image of the object for learning is input to the VAE model that encodes and decodes the input image to output a re-constructed image, and a parameter of the VAE model is adjusted such that a difference between the input image and the re-constructed image decoded from the latent variables is reduced and the distribution of the latent variables in the latent space forms a predetermined probability distribution.
5 . The mark detection method according to claim 4 ,
wherein the step E) includes the steps of:
c) performing machine learning with a first dimensionality being set as a dimensionality of the latent space;
d) changing the dimensionality of the latent space to a second dimensionality smaller than the first dimensionality on the basis of a distribution of the latent variables in each dimension of the latent space, after the pre-adjustment learning step; and
e) performing machine learning with the second dimensionality being set as the dimensionality of the latent space.
6 . The mark detection method according to claim 5 ,
wherein, in the step d), the dimensionality of the latent space is reduced based on variance or standard deviation calculated for each component of the plurality of latent variables in the latent space.
7 . The mark detection method according to claim 4 ,
wherein the step E) includes steps of:
f) performing machine learning based on a first input image;
g) designating a predetermined number of effective components having large standard deviation or variance from among a plurality of components of the latent variables in the latent space, after the first learning step;
h) decoding an effective vector having a value only for the effective component, to generate a corrected image;
i) blending the first input image with the corrected image, to generate a second input image; and
j) performing machine learning on the basis of the second input image.
8 . A computer program that causes the computer to execute the mark detection method according to claim 1 .