A three-dimensional image stitching method for a white light interferometer

Abstract

This invention discloses a three-dimensional image stitching method for a white light interferometer. The sample to be measured is placed on a stage, and a motor moves and collects data in a serpentine path according to a set direction and step distance. Based on the platform movement and data acquisition patterns during large-scale measurements with a white light interferometer, the method automatically determines the data alignment, identifies the optimal feature matching path, and quickly achieves image matching, stitching, and cropping. Furthermore, by utilizing the motor's translational characteristic, the method compares the normalized standard deviation of overlapping regions, automatically searches for the optimal matching region, identifies the actual overlapping region, and completes image stitching. This method features a simple algorithm, low computational load, high speed, and high efficiency. It has low requirements for motor accuracy, low system cost, eliminates the need for manual judgment and selection of feature regions, automatically identifies overlapping regions, and completes image stitching. It boasts a high degree of automation and intelligence, low sample requirements, and is suitable for large-scale measurements of most samples, demonstrating strong versatility.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, specifically to a three-dimensional image stitching method for a white light interferometer. Background Technology

[0002] White light interferometry is one of the most precise techniques for measuring the surface morphology of materials. By determining the position of the coherent envelope peak and combining it with the phase information of the fringes, it can achieve precise measurements at the sub-nanometer level.

[0003] While white-light interferometers offer extremely high measurement accuracy and resolution, their horizontal measurement range is limited by the field of view, making them unsuitable for measuring large-scale horizontal surface topography. Existing methods for achieving large-scale measurements with white-light interferometers can be categorized into two types: one is image stitching based on hardware devices, using external hardware such as motors and grating rulers to achieve high-precision stitching, but this method is costly and lacks system flexibility; the other is matching and stitching based on the geometric features of the sample surface, but existing image stitching methods suffer from complex algorithms, high computational load, slow speed, poor versatility, and low stitching accuracy, failing to adequately meet the needs of large-scale measurements with white-light interferometers.

[0004] Patent 1 (CN105509635A) proposes a method to transform noise into usable information to improve stitching accuracy. In the feature matching stage, noise is transformed into usable features to improve the matching results. In the image fusion stage, the interference of noise regions on data fusion is reduced, resulting in a more accurate 3D stitching result. Targeting the unique functional structure of MEMS, the method utilizes the maximum information entropy principle to solve for event region boundaries in MEMS 3D images, achieving feature segmentation. Then, based on the feature segmentation results, other features besides noise are used for optimized matching and high-level fusion, reducing the impact of noise on 3D image stitching and improving stitching accuracy. Finally, image fusion is performed on the data after noise points are excluded, further reducing the interference of noise regions on data fusion and obtaining a more accurate 3D stitching result.

[0005] The method proposed in Patent 1 transforms noise into usable features to improve the feature matching results. However, the generation of noise is uncertain and may have a negative impact on feature matching. The accuracy of noise recognition has a significant impact on the stitching results and may introduce additional errors, leading to a decrease in stitching accuracy. Furthermore, Patent 1 proposes a method to achieve image stitching by performing feature segmentation and matching on the edge information of the sample surface. However, this method is mainly for large-scale measurements of MEMS and is prone to matching errors for other samples, resulting in low matching accuracy and poor versatility.

[0006] Patent 2 (CN112184551A) proposes a method that combines height quality matrix with 3D image stitching technology to reduce the impact of image noise on 3D stitching. This method first observes data from roughly overlapping areas of two images, identifying a small sub-region as a feature region within this area. Gradient and Harris corner point algorithms are then used to obtain the gradient and corner point matrices of the baseline 3D image. In the roughly overlapping portion of the 3D images to be stitched, the gradient and corner point matrices of each candidate feature region are calculated, and the height quality matrix of that region is obtained using a height quality evaluation method. Next, feature evaluation coefficients are calculated for each candidate feature region and the feature region. Based on the height quality matrix and feature evaluation coefficients, the height of each region is reconstructed. The correlation between the reconstructed height matrix of the feature region and the reconstructed height matrix of the matching region is calculated. The region with the highest correlation coefficient is determined as the optimal matching region. The image to be stitched is moved so that the optimal matching region coincides with the boundary of the feature region of the baseline image. The height of the measured points within the overall area after stitching is fused to the baseline. Combined with the height quality evaluation matrix, an accurate height fusion result without noise is obtained, completing the 3D image stitching.

[0007] The threshold setting of the high-quality evaluation method proposed in Patent 2 lacks a clear standard. The threshold setting is greatly affected by factors such as sample and lighting, which can easily lead to identification errors and a decrease in splicing accuracy. Furthermore, the method proposed in this patent requires manual judgment and selection of feature regions, resulting in poor flexibility and low levels of automation and intelligence.

[0008] Based on the aforementioned problems, this invention proposes a three-dimensional image stitching method for white light interferometers that can solve these problems. Summary of the Invention

[0009] The purpose of this invention is to provide a three-dimensional image stitching method for white light interferometers to solve the problems mentioned in the background art.

[0010] To achieve the above objectives, the present invention provides the following technical solution: a three-dimensional image stitching method for a white light interferometer, comprising the following steps:

[0011] S1. Place the sample to be tested on the stage, set the splicing direction, quantity and motor step distance in the software program, and the motor moves and collects data in a serpentine path according to the set direction and step distance.

[0012] S2. After completing the data acquisition and three-dimensional surface morphology reconstruction of all areas, arrange all three-dimensional image data according to the actual sample image distribution as the image data to be stitched.

[0013] S3. Move the right image to be stitched to the left column by column until the two images gradually overlap. Normalize the data in the overlapping areas and then subtract them to calculate the standard deviation until the last column of the right image overlaps with the left image.

[0014] S4. The overlapping area with the smallest standard deviation is taken as the actual overlapping area of ​​the image data acquired in two adjacent acquisitions.

[0015] S5. Using the left image as a reference, subtract the height data of the overlapping area of ​​the right image from the height data of the overlapping area of ​​the left image, and take the average height difference as the height difference between the two images. Add the calculated height difference to all the data of the right image, and then take the average of the overlapping areas of the left and right images as the value of that pixel. After the image fusion is completed, perform the next column stitching until all columns of images are stitched together.

[0016] S6. After completing the column stitching of all column image data, start row stitching. Move the image below upwards row by row, normalize the data in the overlapping areas, and then subtract them to calculate the standard deviation until the last row of the image below overlaps with the image above.

[0017] S7. The overlapping area with the smallest standard deviation is taken as the actual overlapping area of ​​the image data acquired in two adjacent acquisitions.

[0018] S8. Using the above image as a reference, subtract the height data of the overlapping area of ​​the lower image from the height data of the overlapping area of ​​the upper image, and take the average height difference as the height difference between the two images. Add the calculated height difference to all the data of the lower image, and then take the average of the overlapping areas of the upper and lower images as the value of that pixel. After completing the image fusion, perform the next row stitching until all rows of images are stitched together.

[0019] S9. After completing row stitching, cut off columns with no valid data at the image edges to complete image stitching.

[0020] As a preferred technical solution, in step S6, the number of row splicing times is determined, and if the number of times is odd, the rows are right-aligned; if the number of times is even, the rows are left-aligned.

[0021] As a preferred technical solution, in step S6, for each row moved in the lower image, the data of the currently overlapping area of ​​the lower and upper images are normalized, and then the standard deviation is calculated by subtraction, until the last row of data in the upper and lower images overlaps and the normalization and standard deviation calculation are completed.

[0022] As a preferred technical solution, in step S3, for each column moved in the right image, the data in the currently overlapping area of ​​the left and right images are normalized, and then the standard deviation is calculated by subtracting them, until the last column of data in the left and right images overlaps and the normalization and standard deviation calculation are completed.

[0023] Compared with the prior art, the beneficial effects of the present invention are:

[0024] Based on the platform movement and data acquisition patterns during large-scale measurements with a white light interferometer, it can automatically determine the data alignment method, identify the optimal feature matching path, and quickly achieve image matching, stitching, and cropping. By utilizing the characteristic of the motor only translating, it compares the normalized standard deviation of the overlapping areas, automatically searches for the optimal matching area, identifies the actual overlapping area, and completes image stitching.

[0025] It features simple algorithms, low computational load, high speed, and high efficiency. It has low requirements for motor precision, low system cost, and no need for manual judgment and selection of feature regions. It can automatically identify overlapping regions and complete image stitching. It has a high degree of automation and intelligence, low requirements for samples, and is suitable for large-scale measurement of most samples, with strong versatility. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the serpentine path image acquisition method of the present invention;

[0027] Figure 2 This is a schematic diagram of the column splicing of the present invention;

[0028] Figure 3 This is a schematic diagram of the splicing of the present invention;

[0029] Figure 4 This is a schematic diagram of the image cropping process after stitching together according to the present invention;

[0030] Figure 5 This is a flowchart of the image stitching process of the present invention; Detailed Implementation

[0031] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] Please see Figure 1-5 This invention provides a three-dimensional image stitching method for a white light interferometer:

[0033] The sample to be tested is placed on the stage. The direction, number, and motor step distance are set in the software program. The motor moves in a serpentine path according to the set direction and step distance to collect data and complete the 3D surface topography reconstruction and other post-processing. Figure 1 The diagram shows a serpentine path; the data collection path is indicated by the arrows in the diagram.

[0034] After completing the data acquisition and 3D surface morphology reconstruction of all areas, all 3D image data are arranged in order according to the actual sample image distribution as the image data to be stitched together.

[0035] Because the motor only translates forward, backward, left, and right, the coordinates of the image feature information of the overlapping area in two consecutively acquired images only exhibit translation in a single direction, without multi-directional offset or rotation. Utilizing this characteristic, the feature information of two consecutively acquired images can be compared row by row or column by column according to the direction of translation. The data to be stitched together is then stitched together in column-first, then row-first order. The specific operation is as follows:

[0036] The right image to be stitched is moved column by column to the left, gradually making the two images overlap. Each time the right image moves one column, the data of the current overlapping area of ​​the left and right images is normalized, and then the standard deviation is calculated by subtracting them, until the last column of data of the left and right images overlaps and the normalization and standard deviation calculation are completed. The overlapping area with the smallest standard deviation during this process is taken as the actual overlapping area of ​​the two images.

[0037] Then, using the left image as a baseline, subtract the height data of the overlapping area in the right image from the height data of the overlapping area in the left image, and take the average height difference as the height difference between the two images. Add the calculated height difference to all the data in the right image, and then take the average value of the overlapping area of ​​the left and right images as the value of that pixel. This completes one column stitching. Perform the next column stitching, repeating the above operation until all columns of the image are stitched together. Figure 2 This is a diagram illustrating the splicing of two images. The solid black line represents the overlapping area of ​​the two images, and the arrows indicate the direction of movement.

[0038] After completing column stitching, row stitching begins. Because the motor may have low positioning accuracy or step loss, the length of each row of image data may not be consistent. However, the data of the last captured image in the current row and the first captured image in the next row are aligned. Since the motor moves along a serpentine path, during row stitching, odd-numbered stitches are right-aligned and even-numbered stitches are left-aligned. The lower image to be stitched is moved upwards row by row, gradually making the two images overlap. Each time the lower image moves one row, the data of the currently overlapping area of ​​the lower and upper images is normalized, and the standard deviation is calculated by subtracting them. This process continues until the last row of data of the upper and lower images overlaps and the normalization and standard deviation calculation are completed. The overlapping area with the smallest standard deviation during this process is taken as the actual overlapping area of ​​the two images.

[0039] Then, using the top image as a baseline, subtract the height data of the overlapping area in the bottom image from the height data of the overlapping area in the top image, and take the average height difference as the height difference between the two images. Add the calculated height difference to all the data in the bottom image, and then take the average value of the overlapping area of ​​the top and bottom images as the value of that pixel. This completes one row stitching. Proceed to the next row stitching, repeating the above operation according to the odd-even stitching pattern, until all images are stitched together row by row. Figure 3 This is a diagram illustrating the stitching of images. The solid black line area represents the overlapping area of ​​the two images, and the arrows indicate the direction of movement.

[0040] After row stitching is completed, columns with no valid data in the stitched image are cropped to complete the image stitching, enabling large-scale surface topography measurement, such as... Figure 4 As shown, the dotted line area is the overlapping area of ​​the two images, and the black solid line area is the area retained after cropping.

[0041] Based on the platform movement and data acquisition patterns during large-scale measurements with a white light interferometer, it can automatically determine the data alignment method, identify the optimal feature matching path, and quickly achieve image matching, stitching, and cropping. By utilizing the characteristic of the motor only translating, it compares the normalized standard deviation of the overlapping areas, automatically searches for the optimal matching area, identifies the actual overlapping area, and completes image stitching.

[0042] It features simple algorithms, low computational load, high speed, and high efficiency. It has low requirements for motor precision, low system cost, and no need for manual judgment and selection of feature regions. It can automatically identify overlapping regions and complete image stitching. It has a high degree of automation and intelligence, low requirements for samples, and is suitable for large-scale measurement of most samples, with strong versatility.

[0043] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A three-dimensional image stitching method for a white light interferometer, characterized in that, Includes the following steps: S1. Place the sample to be tested on the stage, set the splicing direction, quantity and motor step distance in the software program, and the motor moves and collects data in a serpentine path according to the set direction and step distance. S2. After completing the data acquisition and three-dimensional surface morphology reconstruction of all areas, arrange all three-dimensional image data according to the actual sample image distribution as the image data to be stitched. S3. Move the right image to be stitched to the left column by column until the two images gradually overlap. Normalize the data in the overlapping areas and then subtract them to calculate the standard deviation until the last column of the right image overlaps with the left image. S4. The overlapping area with the smallest standard deviation is taken as the actual overlapping area of ​​the image data acquired in two adjacent acquisitions. S5. Using the left image as a reference, subtract the height data of the overlapping area of ​​the right image from the height data of the overlapping area of ​​the left image, and take the average height difference as the height difference between the two images. Add the calculated height difference to all the data of the right image, and then take the average of the overlapping areas of the left and right images as the value of that pixel. After the image fusion is completed, perform the next column stitching until all columns of images are stitched together. S6. After completing the column stitching of all column image data, start row stitching. Move the image below upwards row by row, normalize the data in the overlapping areas, and then subtract them to calculate the standard deviation until the last row of the image below overlaps with the image above. S7. The overlapping area with the smallest standard deviation is taken as the actual overlapping area of ​​the image data acquired in two adjacent acquisitions. S8. Using the above image as a reference, subtract the height data of the overlapping area of ​​the lower image from the height data of the overlapping area of ​​the upper image, and take the average height difference as the height difference between the two images. Add the calculated height difference to all the data of the lower image, and then take the average of the overlapping areas of the upper and lower images as the value of that pixel. After completing the image fusion, perform the next row stitching until all rows of images are stitched together. S9. After completing row stitching, cut off columns with no valid data at the image edges to complete image stitching.

2. The three-dimensional image stitching method for a white light interferometer according to claim 1, characterized in that, In step S6, the number of times the rows are spliced ​​is determined. If the number of times is odd, the rows are right-aligned; if the number of times is even, the rows are left-aligned.

3. The three-dimensional image stitching method for a white light interferometer according to claim 2, characterized in that, In step S6, each time the lower image moves one row, the data in the current overlapping area of ​​the lower and upper images are normalized, and then the standard deviation is calculated by subtracting them, until the last row of data in the upper and lower images overlaps and the normalization and standard deviation calculation are completed.

4. The three-dimensional image stitching method for a white light interferometer according to claim 1, characterized in that, In step S3, for each column moved in the right image, the data in the current overlapping area of ​​the left and right images are normalized, and then the standard deviation is calculated by subtracting them, until the last column of data in the left and right images overlaps and the normalization and standard deviation calculation are completed.

Images