JP2022163862A - Image processing method and image processing device - Google Patents

Abstract

To appropriately reduce noise in various captured images such as images obtained by scanning an imaging target with a charged particle beam.SOLUTION: An image processing method includes the steps of: (A)repeating phase correction multiple times to correct a phase of transformed data obtained by Fourier transforming an original image with a correction amount randomly determined according to a frequency for each frequency component; (B) generating a plurality of artificial images by inversely Fourier transforming each of the multiple pieces of phase corrected transformed data; and (C) averaging the plurality of artificial images to generate an artificial average image.SELECTED DRAWING: Figure 8

Description

The present disclosure relates to image processing methods and image processing apparatuses.

Patent Document 1 discloses a method of obtaining an image by scanning a pattern on a wafer with an electron beam, in which an image with a high S/N ratio is formed by integrating signals obtained in a plurality of frames. is disclosed.

JP 2010-92949 A

A technique according to the present disclosure appropriately reduces noise in various captured images such as an image obtained by scanning an imaging target with a charged particle beam.

According to one aspect of the present disclosure, (A) phase correction is performed by correcting the phase of transformed data obtained by Fourier transforming an original image with a correction amount randomly determined according to the frequency for each frequency component. (B) performing an inverse Fourier transform on each of the phase-corrected transformed data to generate a plurality of artificial images; and (C) the plurality of artificial images. averaging the artificial images of to generate an artificial average image.

According to the present disclosure, it is possible to appropriately reduce noise in various captured images such as images obtained by scanning an imaging target with a charged particle beam.

FIG. 10 is a diagram showing the result of PSD (Power Spectral Density) analysis of a frame image obtained by scanning a patterned substrate with an electron beam; FIG. 10 illustrates the effect of the phase of each of the low frequency pre-high frequency waves on a composite wave of a high amplitude low frequency wave and a low amplitude high frequency wave. FIG. 4 is a diagram showing low frequency components of a frame image; FIG. 4 is a diagram showing high frequency components of a frame image; 1 is a diagram showing a schematic configuration of a processing system including a control device as an image processing device according to the first embodiment; FIG. 3 is a block diagram showing an outline of a configuration related to image processing of a control unit; FIG. 4 is a flowchart for explaining an example of processing in a control unit; FIG. 4 is a conceptual diagram showing an example of the flow of processing in a control unit; FIG. 4 is a conceptual diagram showing an example of phase correction; FIG. 4 is a comparison diagram of one actual frame image and an artificial average image generated from the one actual frame image by the method according to the present embodiment; FIG. 5 is a comparison diagram of an average image obtained by averaging four actual frame images and an artificial average image generated from the four actual frame images by the method according to the present embodiment; FIG. 7 is a conceptual diagram showing another example of the flow of processing in the control unit; FIG. 7 is a conceptual diagram showing another example of the flow of processing in the control unit; FIG. 7 is a conceptual diagram showing another example of the flow of processing in the control unit; FIG. 10 is a diagram for explaining another example of correction performed by the correction unit;

2. Description of the Related Art An image obtained by scanning a substrate with a charged particle beam (for example, an electron beam) is used for inspection, analysis, etc. of a fine pattern formed on a substrate during the manufacturing process of a semiconductor device or the like. Images used for inspection, analysis, etc. are required to have little noise.
In Patent Literature 1, an image with a high S/N ratio, that is, an image with little noise is formed by integrating signals acquired in a plurality of frames.

By the way, in recent years, further miniaturization of semiconductor devices is required. Along with this, there is a demand for further noise reduction in images used for pattern inspection, analysis, and the like.
In addition, there is a demand for further reduction of noise for objects to be imaged other than the substrate.
Further reduction of noise is a common problem for captured images other than images obtained by scanning with a charged particle beam.

Therefore, the technology according to the present disclosure appropriately reduces noise in various captured images such as images obtained by scanning an imaging target with a charged particle beam.

<Circumstances leading to the technology according to the present disclosure>
A frame image obtained by scanning an observation area of an imaging target with a charged particle beam such as an electron beam includes image noise caused by imaging conditions and imaging environment. Image noise must be removed from images used for analysis and the like.

Techniques for removing image noise include the following. That is, this is a technique in which high-frequency components obtained by Fourier transforming a frame image are removed as image noise using a low-pass filter, and the data after the removal is subjected to inverse Fourier transform to obtain an image used for analysis or the like. However, since the high-frequency component includes not only image noise but also signals that should not be removed, the above method cannot properly remove image noise.

In the fields of voice and music, although not in the field of images, noise is removed (cancelled) by synthesizing a noise component and an antiphase component.
As for images, even if high-frequency components and anti-phase components are synthesized in the same manner as voice/music, image noise cannot be removed appropriately because the high-frequency components also contain signals as described above. If the phases of the synthesized high-frequency components are not reversed, but are adjusted appropriately, it may be possible to remove only the image noise components of the high-frequency components, but the image noise will fluctuate between frames. In view of the above, it is unrealistic to appropriately adjust the phase of the high frequency component to be synthesized for each frame.

High-frequency components in a frame image include a signal-derived component with phase φ1 and an image-noise-derived component with phase φ2. It is assumed that the phase φ2 of the derived component varies randomly and differs from frame to frame.
Therefore, if a large number of frame images (for example, 32 times) are averaged, among the high-frequency components, image noise-derived components cancel each other out, so that only the signal-derived component with phase φ1 remains. However, when the number of frames is increased, the charged particle beam damages the object to be imaged.

On the other hand, as shown in the results of PSD (Power Spectral Density) analysis in FIG. 1, when a substrate on which a pattern is formed is an object to be imaged, the frame image has lower high-frequency components when subjected to Fourier transform. It turns out that the amplitude (intensity is the square of the amplitude) is small compared to the frequency component.

Here, consider a composite wave C of a low-frequency wave L having a large amplitude and a high-frequency wave H having a small amplitude, as shown in FIG. As shown in FIGS. 2(a) and 2(b), before and after the phase of the low-frequency wave L with large amplitude is shifted, the shape of the composite wave C changes greatly. As shown in FIGS. 2(a) and 2(c), the shape of the composite wave C does not change significantly before and after the phase of the wave H is shifted.

Applying these to the data obtained by Fourier transforming the frame image results in the following. That is, as shown in FIG. 3, when the phase φ of the low-frequency component with a large amplitude r is shifted, the effect on the entire Fourier-transformed data is large. , is considered to be greatly different from the original frame image. On the other hand, even if the phase φ of the high-frequency component with a large amplitude r is shifted as shown in FIG. The image is assumed to be nearly identical to the original frame image.

If a large number of images obtained by performing inverse Fourier transform after randomly shifting the phase φ are prepared and averaged, the following effects can be expected. That is, similar to the method of averaging a large number (for example, 32) of actual frame images, among the high-frequency components, components derived from image noise whose phase φ2 changes randomly between frames in the actual frame images are removed. , only the components derived from signals having a common phase φ1 between frames can be left in an actual frame image. In addition, a large number of images obtained by performing inverse Fourier transform after randomly shifting the phase φ used in this method can be prepared even if the number of times of scanning the imaging target with the charged particle beam is small. Therefore, it is possible to suppress damage to the imaging target due to the charged particle beam.

The present embodiment is based on the above study results.

An image processing method and an image processing apparatus according to the present embodiment will be described below with reference to the drawings. In this specification, elements having substantially the same functional configuration are denoted by the same reference numerals, thereby omitting redundant description.

<Processing system>
FIG. 5 is a diagram showing a schematic configuration of a processing system including a control device as an image processing device according to this embodiment.
The processing system 1 of FIG. 5 has a scanning electron microscope 10 and a controller 20 .

A scanning electron microscope 10 has an electron source 11 , a deflector 12 and a detector 13 .

The electron source 11 emits electron beams as charged particle beams.
The deflector 12 is for two-dimensionally scanning an imaging region of a semiconductor wafer (hereinafter referred to as a wafer) W as a patterned substrate with an electron beam from the electron source 11 .
The detector 13 amplifies and detects secondary electrons generated from the wafer W by irradiation of the electron beam.

The control device 20 has a storage section 21 , a display section 22 and a control section 23 .

The storage unit 21 stores various types of information, and includes a memory such as a RAM (Random Access Memory) and a storage device such as a HDD (Hard Disk Drive). The storage unit 21 stores programs and the like for image processing in the control unit 23 . The program may be recorded in a computer-readable storage medium M and installed in the control device 20 from the storage medium M. Moreover, the storage medium M may be temporary or non-temporary.

The display unit 22 displays various information, and is configured by a display device such as a liquid crystal display or an organic display.

The controller 23 controls the scanning electron microscope 10 and the controller 20 . The control unit 23 has a processor such as a CPU, for example.

<Control part>
FIG. 6 is a block diagram showing an outline of a configuration related to image processing of the control section 23. As shown in FIG.
As shown in FIG. 6, the control unit 23 includes a frame image generation unit 201, an acquisition unit 202, and a Fourier transform, which are implemented by a processor such as a CPU reading and executing a program stored in the storage unit 21. It has a section 203 , a correction section 204 , an inverse Fourier transform section 205 , an image generation section 206 and a measurement section 207 .

The frame image generation unit 201 sequentially generates frame images of the wafer W having patterns based on the detection results of the detector 13 of the scanning electron microscope 10 . The frame image generation unit 201 generates a specified number of frame images (for example, four) of the wafer W having the pattern, that is, the specified number of frames. Also, the generated frame images are sequentially stored in the storage unit 21 .

The acquisition unit 202 acquires the frame image generated by the frame image generation unit 201 and stored in the storage unit 21 .

A Fourier transform unit 203 Fourier transforms the original image to generate transformed data. In this embodiment, the Fourier transform unit 203 Fourier transforms the frame image acquired by the acquisition unit 202 to generate transformed data.

A correction unit 204 performs phase correction on the transformed data generated by the Fourier transform unit 203 . The phase correction here means correcting the phase φ with a correction amount randomly determined according to the spatial frequency for each spatial frequency component. For example, the correction amount of the phase φ in the phase correction is randomly determined from a predetermined range (hereinafter referred to as a correctable range) for each spatial frequency. In addition, the absolute values of the upper and lower limits of the correctable range are set larger when the spatial frequency is high than when the spatial frequency is low. In other words, the correctable range of low spatial frequencies is included in the correctable range of high spatial frequencies. Specifically, for example, the correctable range for the highest spatial frequency is set to -π to +π, and the correctable range for the lowest spatial frequency is set to zero.

An inverse Fourier transform unit 205 performs an inverse Fourier transform on each of the plurality of transformed data phase-corrected by the corrector 204 to generate a plurality of artificial images.

The image generation unit 206 averages the plurality of artificial images generated by the inverse Fourier transform unit 205 to generate an artificial average image.

The measurement unit 207 measures the feature amount of the pattern of the wafer W based on the artificial average image generated by the image generation unit 206 . The pattern feature amount is, for example, the edge coordinates of the pattern.

<Processing example 1 by control unit>
FIG. 7 is a flowchart for explaining an example of processing in the control unit 23. As shown in FIG. FIG. 8 is a conceptual diagram showing an example of the flow of processing in the control unit 23. As shown in FIG. FIG. 9 is a conceptual diagram showing an example of phase correction. In the following process, the scanning electron microscope 10 is controlled by the control unit 23 to perform scanning with an electron beam for P (≧1) frames specified by the user, that is, P frames. It is assumed that the specified P frame images have already been generated. It is also assumed that the generated frame images are stored in the storage unit 21 .

In the processing in the control unit 23, first, the acquisition unit 202 acquires P frame images F from the storage unit 21, as shown in FIGS. 7 and 8 (step S1). Note that when P≧2, the image size and imaging area are common among the acquired frame images F. Also, the size of the acquired frame image is, for example, 1000×1000 pixels, and the size of the imaging area is 2000 nm to 100 nm square.

Next, the Fourier transform unit 203 performs two-dimensional Fourier transform on the P frame images F acquired by the acquisition unit 202 to create transformed data D (step S2).
Specifically, when there are a plurality of frame images F acquired by the acquisition unit 202, that is, when P≧2, the Fourier transform unit 203 converts the P (≧2) frame images F acquired by the acquisition unit 202 are averaged to generate an average image FA, and the average image FA is subjected to a two-dimensional Fourier transform to create transformed data D.
When the number of frame images F acquired by the acquisition unit 202 is one, that is, when P=1, the Fourier transform unit 203 performs two-dimensional Fourier transform on the frame image F acquired by the acquisition unit 202. Create data D.
In the two-dimensional Fourier transform, Fourier coefficients a and b are calculated for each spatial frequency in each of the horizontal and vertical directions, and the amplitude r and phase φ are calculated from the Fourier coefficients a and b.

Subsequently, the correction unit 204 performs the above-described phase correction on the transform data D generated by the Fourier transform unit 203, that is, for each spatial frequency component of the transform data D, Correction of the phase by a randomly determined correction amount is repeated Q (Q is greater than P, for example, 32) times (step S3).

Specifically, the correcting unit 204 repeats Q times correcting the phase φ of each frequency component of the transform data D with a correction amount randomly determined from a correctable range preset for each spatial frequency.

More specifically, as shown in FIG. 9, the correction unit 204 uses, for example, a phase φ template T1 and a random number template T2 for phase correction.
In the template T1 of the phase φ, the absolute values of the upper and lower limits of the correctable range are set for each coordinate in the spatial frequency domain. ing. For the template T1 of the phase φ, for example, the absolute value of the highest spatial frequency fma is set to π, the absolute value of the lowest spatial frequency fmi is set to zero, and the spatial frequency between fma and fmi is set to π. is set so as to increase in two or more steps or to increase monotonically as the spatial frequency increases. The reason why the absolute value is increased with an increase in the spatial frequency is that the higher the spatial frequency, the more likely the image is affected by image noise, and the greater the need for phase correction.
The random number template T2 is set with a random number generated by a random number generator (not shown) for each coordinate in the spatial frequency domain. set.

The correction unit 204 determines a correction amount for phase correction of the transformed data D based on the phase φ template T1 and the random number template T2 for each spatial frequency component. For example, for each spatial frequency component, the correction unit 204 extracts the absolute value associated with the corresponding frequency from the template T1 of the phase φ, and extracts the random number associated with the corresponding frequency from the template T2. Then, the correction unit 204 multiplies the extracted absolute value by the extracted random number for each spatial frequency component, determines the result as a correction amount, and adds the determined correction amount to the phase.
That is, the correction unit 204 multiplies the template T1 of the phase φ by the random number template T2 to generate the template T3 of the correction amount of the phase φ, and adds the template T3 of the correction amount of the phase φ to the conversion data D.
In step S4, the correction unit 204 repeatedly generates the random number template T2 Q times, and performs the above-described processing on the conversion data D each time the template T2 is generated.

Next, as shown in FIGS. 7 and 8, the inverse Fourier transform unit 205 inverse Fourier transforms the Q pieces of transformed data D that have been phase-corrected by the correction unit 204, and generates Q artificial images J. Generate (step S4). In the inverse Fourier transform, Fourier coefficients a' and b' are calculated based on the amplitude r and the corrected phase φ' for each spatial frequency in the horizontal direction and the vertical direction. , an artificial image J is generated.
Steps S4 and S3 are repeated by the number of artificial average images JA generated in the subsequent step S5.

Subsequently, the image generator 206 averages the Q artificial images J generated by the inverse Fourier transform unit 205 to generate an artificial average image JA (step S5). In this step S5, the image generation unit 206 generates, for example, R artificial average images JA designated by the user.

Then, the measurement unit 207 measures the edge coordinates of the pattern as the feature amount of the pattern of the wafer W based on the artificial average image JA generated by the image generation unit 206 (step S6).
Specifically, the measurement unit 207 measures the edge coordinates of the pattern for each of the R artificial average images JA generated by the image generation unit 206, and calculates the average value of the edge coordinates as the statistic value of the edge coordinates. calculate.
The artificial average image JA may be displayed on the display unit 22 at the same time as, before, or after the measurement of the edge coordinates.

Thereby, a series of processes in the control unit 23 is completed.

<Method of averaging>
In this embodiment, when the artificial average image JA is generated from Q artificial images J, various values are averaged when calculating the average value of the edge coordinates.
As a method for this averaging, a method of averaging after removing (excluded) outliers may be adopted.
Specifically, in step S5, the image generating unit 206 removes outliers in pixel values from the Q artificial images J, and then averages the Q artificial images J to generate an average artificial image JA. may
Further, in step S6, the measurement unit 207 may calculate (acquire) the average value of the edge coordinates after removing outliers of the edge coordinates. Outliers may be removed (excluded) when calculating (acquiring) statistics other than the average value.

When P (≧2) frame images F are averaged, even if the number of frame images F is large, the Fourier transform unit 203 similarly removes outliers of pixel values from the P frame images F. After that, the P frame images F may be averaged to generate the average image FA.

In addition, the said outlier is identified using a well-known method. For example, the outliers are identified based on the Smirnov-Grubbs test.

Also, the outliers may be specified using a standard score such as a Z score. Note that the Z score z for the variable x is calculated based on the following formula using the average value μ and the standard deviation σ.
z=(x−μ)/σ

Further, the outliers may be identified using values of surrounding similarities (eg, neighboring pixels).

<Effects of this embodiment>
As described above, in the present embodiment, the phase φ of the high-frequency component and the like are applied to the average image FA of P frame images F or the transformed data D obtained by two-dimensional Fourier transforming one frame image F. Randomly shifted phase correction is repeated Q times, that is, a large number of times. Then, in the present embodiment, Q, i.e., a large number of transformed data D after phase correction are subjected to an inverse Fourier transform, Q, i.e., a large number of artificial images J are generated, and these are averaged to generate an artificial average image JA. there is Therefore, of the high-frequency components of the artificial average image JA, only components derived from image noise whose phase changes randomly between frames in the actual frame image are removed. The component derived from the signal with is left without being removed. That is, according to this embodiment, it is possible to generate an artificial average image JA from which image noise is appropriately removed.

In addition, in order to appropriately remove image noise from the artificial average image JA, it is necessary to increase the number of artificial images J constituting the artificial average image JA. A large number of artificial images J can be prepared even if the number of actual frame images is small. For example, even if the actual number of frame images is as small as 1 or 4, a large number of artificial images J such as 32 or more can be prepared. Therefore, according to this embodiment, it is possible to appropriately remove image noise while suppressing damage to the wafer W due to the electron beam.

(Example)
FIG. 10 is a comparison diagram of one actual frame image F and an artificial average image JA generated from one actual frame image F by the method according to this embodiment. FIG. 11 shows an average image (hereinafter referred to as simple average image) FA obtained by averaging four actual frame images F, and an artificial average image JA generated from the four actual frame images F by the method according to the present embodiment. It is a comparison diagram with. The imaging target of the frame image F is the wafer W on which the cylindrical pattern is formed, and the size of the imaging area is 250 nm square. Each artificial average image JA is obtained by averaging 32 artificial images J. FIG.

The artificial average image JA in FIG. 10 has a clearer contour of the cylindrical pattern than the single actual frame image F, that is, the contrast is higher. Also, in the artificial average image JA of FIG. 10, the brightness of the bright spots, which are image noise, is lower than the brightness of the pattern portion around the cylindrical pattern compared to the one actual frame image F. Also, the number of bright spots, which are image noise, is reduced.
Similarly, in the artificial average image JA of FIG. 11, the outline of the cylindrical pattern is clearer than in the simple average image FA. Also, in the simple average image FA, there are bright spots, which are image noise, around the cylindrical pattern, but in the artificial average image JA of FIG. 11, there are almost no bright spots.
From the results shown in FIGS. 10 and 11, it can be seen that the artificial average image JA from which the image noise is appropriately removed can be generated according to the present embodiment.

<Processing example 2 by the control unit>
FIG. 12 is a conceptual diagram showing another example of the flow of processing in the control unit 23. As shown in FIG.
In the processing example 1 described above, when there are a plurality of frame images F acquired by the acquisition unit 202, that is, when P≧2, the Fourier transform unit 203 converts P (≧2) frames acquired by the acquisition unit 202 into The image F is averaged to generate an average image FA, and the average image FA is two-dimensionally Fourier-transformed to create the transformed data D.

Alternatively, as shown in FIG. 12, the Fourier transform unit 203 performs a two-dimensional Fourier transform on each of the P (≧2) frame images F acquired by the acquisition unit 202 to obtain transformed data D may be created.
In this case, the correction unit 204 performs phase correction on the conversion data D of each of P (≥2) frame images F, that is, a plurality of frame images F, by Q (Q is larger than P, for example, 32 ) times, i.e. multiple times. In other words, the correction unit 204 performs the above-described phase correction on each of the P pieces of transformed data D generated by the Fourier transform unit 203 by Q (Q is larger than P, for example, 32 ) times.

For each of the P frame images F, the inverse Fourier transform unit 205 generates Q (Q is larger than P, for example, 32) artificial images, that is, a plurality of artificial images. In other words, the inverse Fourier transform unit 205 performs an inverse Fourier transform on each of the P×Q pieces of transformed data D phase-corrected by the correcting unit 204 to generate P×Q artificial images J. FIG.

Then, the image generator 206 collectively averages Q (Q is larger than P, for example, 32), i.e., a plurality of artificial images J generated for each of P (≧2), i.e., a plurality of frame images F. Generate an artificial average image JA. In other words, the image generator 206 mixes and averages the P×Q artificial images J generated by the inverse Fourier transform unit 205 to generate the artificial average image JA.

By generating a plurality of artificial images J based on a plurality of types of conversion data D as in this processing example 2, a greater variety of artificial images J can be generated. Therefore, by averaging these artificial images J to generate an artificial average image JA, image noise can be removed more appropriately than in the processing example 1 described above.

<Processing example 3 by control unit>
FIG. 13 is a conceptual diagram showing another example of the flow of processing in the control unit 23. As shown in FIG.
In this processing example, as shown in FIG. 13, the Fourier transform unit 203 transforms each of the P (≧2) frame images F acquired by the acquisition unit 202 into Transformed data D is created by two-dimensional Fourier transformation.

However, unlike the processing example 2 described above, the correction unit 204 converts P (≧2) pieces of transformed data D obtained by two-dimensional Fourier transforming each of P (≧2) frames F, that is, a plurality of frame images F to It analyzes and adjusts the correction amount in the phase correction based on the analysis result. Then, for example, the correction unit 204 performs phase correction in which the correction amount is adjusted for one transformation data D obtained by Fourier transforming one image out of the P (≧2) frame images F. is repeated Q (Q is greater than P, for example, 32) times (that is, multiple times).

Specifically, the correction unit 204 corrects the template T1 of the phase φ based on the result of the variation analysis of the P (≧2) pieces of the transformed data D. For example, the correction unit 204 identifies, from the P (≧2) pieces of transformed data D, a frequency with a large phase change between the frame images F. and the template T1 of the phase φ is adjusted so that the absolute value of the lower limit is increased.

Then, the correction unit 204 determines a correction amount for each spatial frequency component based on the adjusted phase φ template T1′ and the random number template T2. Then, the correction unit 204 determines one conversion data D obtained by Fourier transforming, for example, the top one image among the P (≧2) frame images F as described above. Add the correction amount to the phase. The correction unit 204 repeatedly generates the random number template T2 Q times, and adds the correction amount determined as described above to the phase of the conversion data D each time it is generated.

Note that the processing performed by the inverse Fourier transform unit 205 and the image generation unit 206 is the same as in the processing example 1 described above.

By generating the artificial average image JA as in Processing Example 3, image noise can be removed more appropriately than in Processing Example 1 described above.

<Processing example 4 by control unit>
FIG. 14 is a conceptual diagram showing another example of the flow of processing in the control unit 23. As shown in FIG.
In this processing example, as shown in FIG. 14, unlike the processing examples 1 to 3 described above, the Fourier transform unit 203 performs P (≧2) frame images F acquired by the acquisition unit 202, that is, time-series frame images. FG is subjected to a three-dimensional Fourier transform, and P pieces of transformed data D are created.

Then, the correction unit 204 repeats the phase correction of the transformed data D obtained by three-dimensional Fourier transforming the time-series frame images FG Q times (Q is greater than P, for example, 32), that is, a plurality of times. .
Specifically, the correction unit 204 performs phase correction on n pieces of transformed data D obtained by Fourier transforming n frame images F (for example, the first frame image F) that constitute the time-series frame images FG. Correction is repeated Q (Q is greater than P, eg, 32)/n times.

Note that the processes performed by the inverse Fourier transform unit 205 and the image generation unit 206 are the same as those in the above-described processing example 1 and the like.

Image noise can also be appropriately removed by generating the artificial average image JA as in this processing example 4. FIG.

<Another example of the original image transformed by the Fourier transform unit>
In the above example, the original image to be Fourier-transformed by the Fourier transform unit 203 was an actual frame image obtained by scanning the object to be imaged with a charged particle beam such as an electron beam.
The original image is another artificial image generated by the method described in WO 2020/090206 when P≧2 in Modification 1, Modification 2, or Modification 3, and good too.

The other artificial image is generated, for example, as follows.
That is, first, the acquisition unit 202 acquires a plurality of (for example, four) actual frame images F from the storage unit 21 . Next, another image generation unit (not shown) included in the control unit 23 determines the luminance probability distribution for each pixel from the plurality of actual frame images. Then, the another image generating unit generates P (≧2) different artificial images based on the probability distribution of luminance for each pixel.

Note that the above another image generation unit is realized by reading and executing a program stored in the storage unit 21 by a processor such as a CPU of the control unit 23 .

Further, in the case of Modifications 1 to 3, the original image may be an artificial average image JA that has been processed and created in the past as in Modifications 1 to 3. FIG.

Furthermore, the original image may be an image obtained by applying a known noise filter such as a box filter to the actual frame image.

Further, the original image may be an image captured by an imaging device having a light receiving element. In this case, it is not limited to a monochrome image such as a black and white image, and may be a multicolor image such as an RGB image. In the case of a multicolor image, Fourier transform, phase correction, and the like are performed, for example, for each single color image forming the multicolor image, or only for a specific single color image.

<Modified example of correction performed by correction unit>
FIG. 15 is a diagram for explaining another example of correction performed by the correction unit 204. FIG.
In the above example, the correction unit 204 performs only phase correction on the transformed data D, but as shown in FIG. Additionally, a correction of the amplitude r may be performed. That is, the correction unit 204 may perform amplitude correction for correcting the amplitude r by a correction amount randomly determined according to the frequency for each frequency component together with the phase correction. In this case, the correction amount of the amplitude r in the amplitude correction is randomly determined from a predetermined range for each frequency, similar to the phase φ. The absolute value of the lower limit is set to be large. For example, the absolute value for the highest spatial frequency fma is the amplitude r×100% obtained by the Fourier transform.

<Other examples of phase templates>
In the above example, in the template T1 of the phase φ, the absolute values of the upper and lower limits of the correctable range for the lowest spatial frequency fmi were set to zero. The absolute value may not be zero as long as there is no influence.

It should be considered that the embodiments disclosed this time are illustrative in all respects and not restrictive. The embodiments described above may be omitted, substituted, or modified in various ways without departing from the scope and spirit of the appended claims.

20 Control device 204 Correction unit 205 Inverse Fourier transform unit 206 Image generation unit D Conversion data F Frame image FA Average image FG Time series frame image J Artificial image JA Artificial average image φ Phase

Claims (20)

(A) A step of repeating a plurality of times to correct the phase of the transformed data obtained by Fourier transforming the original image with a correction amount randomly determined according to the frequency for each frequency component. When,
(B) generating a plurality of artificial images by inverse Fourier transforming each of the plurality of phase-corrected transformed data;
(C) An image processing method, comprising the step of averaging the plurality of artificial images to generate an artificial average image. The step (A) randomly determines a correction amount in the phase correction from a predetermined range for each frequency,
2. The image processing method according to claim 1, wherein the absolute values of the upper and lower limits of the range are larger when the frequency is high than when the frequency is low. The step (A) includes a phase template in which the absolute values of the upper limit and the lower limit are set for each coordinate in the spatial frequency domain, and a random number template in which a random number is set for each coordinate in the spatial frequency domain. 3. The image processing method according to claim 2, wherein the correction amount in said phase correction is determined based on . the original image is an image of a patterned substrate;
4. The image processing method according to any one of claims 1 to 3, further comprising the step of (D) measuring the feature amount of said pattern based on said artificial average image. 5. The image processing method according to claim 4, wherein the feature amount of said pattern is edge coordinates of said pattern. The step (C) generates a plurality of the artificial average images,
6. The method according to claim 4, wherein the step (D) measures the feature amount of the pattern based on each of the artificial average images, and acquires the statistic of the feature amount of the pattern based on the measurement result. Image processing method. 7. The image processing method according to claim 6, wherein said statistic is an average value. 8. The image processing method according to claim 6, wherein said step (D) obtains a statistic from which outliers are removed from said pattern feature amount. The image processing method according to any one of claims 1 to 8, wherein the step (C) averages the plurality of artificial images from which outliers of pixel values are removed to generate the artificial average image. The image processing method according to any one of claims 1 to 9, wherein said original image is an image obtained by averaging a plurality of captured images. The image processing method according to any one of claims 1 to 9, wherein said original image is a single captured image. In the step (A), for each of the plurality of captured images, performing the phase correction on the transformed data obtained by two-dimensional Fourier transforming the captured image is repeated multiple times,
The step (B) generates the plurality of artificial images for each of the plurality of captured images,
12. The image processing method according to claim 11, wherein in step (C), the plurality of artificial images generated for each of the plurality of captured images are collectively averaged to generate the artificial average image. The step (A) is a step of analyzing a plurality of transformed data obtained by performing a two-dimensional Fourier transform on each of the plurality of captured images, and adjusting the correction amount in the phase correction based on the analysis result. 12. The image processing according to claim 11, wherein performing the phase correction with the adjusted correction amount on the transformed data obtained by Fourier transforming the single captured image is repeated a plurality of times. Method. The original image is a time-series captured image,
10. The method according to any one of claims 1 to 9, wherein in the step (A), the phase correction is repeated a plurality of times for transformed data obtained by performing a three-dimensional Fourier transform on the time-series captured images. image processing method. 15. The image processing method according to any one of claims 10 to 14, wherein said captured image is an actual frame image obtained by scanning an object to be imaged with a charged particle beam. (E) acquiring a plurality of actual frame images obtained by scanning the imaging target with the charged particle beam;
(F) determining a luminance probability distribution for each pixel from the plurality of actual frame images;
(G) generating a plurality of different artificial images based on the probability distribution of luminance for each pixel;
14. The image processing method according to any one of claims 10, 12 and 13, wherein said captured image is another artificial image generated in said step (G). 14. The image processing method according to any one of claims 10 to 13, wherein said captured image is said artificial average image generated in said step (C) in the past. 18. The step (A) of any one of claims 1 to 17, wherein the step (A) performs amplitude correction for correcting the amplitude of the transformed data by a correction amount randomly determined according to the frequency for each frequency component, together with the phase correction. 1. The image processing method according to 1. A readable computer storage storing a program that runs on a computer of a control unit that controls an image processing apparatus so that the image processing method according to any one of claims 1 to 18 is executed by the image processing apparatus. medium. a correcting unit that repeats a number of times to correct the phase of the transformed data obtained by Fourier transforming the original image with a correction amount that is randomly determined according to the frequency for each frequency component; ,
an inverse Fourier transform unit for inverse Fourier transforming the plurality of phase-corrected transformed data to generate a plurality of artificial images;
and an image generator that averages the plurality of artificial images to generate an artificial average image.

Images