Image correcting method
View Patent ↗In a reticle inspecting apparatus or the like, there is provided an image correcting method which is effective when a rank of a matrix lacks due to continuous equal grayscale values when an image is handled as a matrix. In the image correcting method, a random noise image having fine grayscale is superposed on a pattern image to make a matrix full-rank.
1. An image correcting method for generating a correction image from 1st and 2nd pattern images, said method comprising:
generating a random noise pattern image having grayscale values that are finer than grayscale values of the 1st pattern image;
superposing the random noise pattern image at least on a region of the 1st pattern image having equal grayscale values within the region;
identifying a linear prediction model that describes an input-output relationship of the images by operations of: (1) forming a relational expression of a matrix using the 2nd pattern image as an output, and using the 1st pattern image on which the random noise is superposed as an input, (2) generating simultaneous equations from the matrix, (3) solving the simultaneous equations to estimate parameters of the linear prediction model; and
generating a correction image by applying the linear prediction model with the estimated parameters to the 1st pattern image.
2. An image correcting method for generating a correction image from an inspection reference pattern image and a pattern image under test, said method comprising:
generating a random noise pattern image having grayscale values that are finer than grayscale values of the inspection reference pattern image;
superposing the random noise pattern image at least on a region of the inspection reference pattern image having equal grayscale values within the region;
identifying a linear prediction model that describes an input-output relationship of the images by operations of: (1) forming a relational expression of a matrix using the pattern image under test as an output, and using the inspection reference pattern image on which the random noise is superposed as an input, (2) generating simultaneous equations from the matrix, (3) solving the simultaneous equations to estimate parameters of the linear prediction model; and
generating a correction image by applying the linear prediction model with the estimated parameters to the inspection reference pattern image.
3. An image correcting method for generating a correction image from an inspection reference pattern image and a pattern image under test, said method comprising:
generating two random noise pattern images having grayscale values that are finer than grayscale values of the inspection reference pattern image and the pattern image under test; and
superposing the two random noise pattern images at least on a first region of the inspection reference pattern image having equal grayscale values within the first region and on a second region of the pattern image under test having equal grayscale values within the second region, respectively;
identifying a linear prediction model that describes an input-output relationship of the images by operations of: (1) forming a relational expression of a matrix using the pattern image under test as an output, and using the inspection reference pattern image on which the random noise is superposed as an input, (2) generating simultaneous equations from the matrix, (3) solving the simultaneous equations to estimate parameters of the linear prediction model; and
generating a correction image by applying the linear prediction model with the estimated parameters to the inspection reference pattern image.
4. An image correcting method for generating a correction image from an inspection reference pattern image and a pattern image under test, said method comprising:
setting regions within the inspection reference pattern image and the pattern image under test as uninspected regions;
setting minimum grayscale values on the uninspected regions;
generating two random noise pattern images having grayscale values that are finer than grayscale values of the inspection reference pattern image and the pattern image under test;
superposing the two random noise pattern images on the uninspected regions of the inspection reference pattern image and the pattern image under test, respectively;
identifying a linear prediction model that describes an input-output relationship of the images by operations of: (1) forming a relational expression of a matrix using the pattern image under test on which the random noise is superposed as an output, and using the inspection reference pattern image on which the random noise is superposed as an input, (2) generating simultaneous equations from the matrix, (3) solving the simultaneous equations to estimate parameters of the linear prediction model; and
generating a correction image by applying the linear prediction model with the estimated parameters to the inspection reference pattern image.
5. An image correcting method for generating a correction image from an inspection reference pattern image and a pattern image under test, said method comprising:
generating a random noise pattern image having grayscale values that are finer than grayscale values of the inspection reference pattern image;
superposing the random noise pattern image at least on a region of the inspection reference pattern image having equal grayscale values within the region;
generating simultaneous equations which describe an input-output relationship using, as an output, each pixel of the pattern image under test and using, as an input, a linear coupling of a pixel group of the reference pattern image on which the random noise is superposed around each pixel corresponding to the each pixel of the pattern image under test of on which the random noise is superposed;
solving the simultaneous equations to estimate parameters of a prediction model; and
generating a correction image by using the prediction model with the estimated parameters.
6. The image correcting method according to claim 5 , wherein
the prediction model is a two-dimensional prediction model using each pixel of the pattern image under test as two-dimensional output data and using a linear coupling of a pixel group around each pixel as two-dimensional input data.
7. The image correcting method according to claim 5 , wherein
the parameters of the prediction model are estimated by using a least-square method.
8. An image correcting method for generating a correction image from an inspection reference pattern image and a pattern image under test, said method comprising:
generating a random noise pattern image having grayscale values that are finer than grayscale values of the inspection reference pattern image;
superposing the random noise pattern image at least on a region of the inspection reference pattern image having equal grayscale values within the region;
setting reference points at a plurality of separated positions in the inspection reference pattern image on which the random noise is superposed, giving a weight to the inspection reference pattern image on which the random noise is superposed with reference to the reference points, and generating decomposed images the number of which is equal to the number of reference points;
generating simultaneous equations which describe an input-output relationship using each pixel of the pattern image under test on which the random noise is superposed as an output and using a linear coupling of a pixel group of the decomposed image around each pixel corresponding to the each pixel of the pattern image under test as an input;
solving the simultaneous equations to estimate parameters of a prediction model; and
generating a correction image by using the prediction model with the estimated parameters.