A method for extracting a fluorescent boundary line of a laser-induced transparent ice body emitting fluorescence

Abstract

This invention provides a method for extracting the fluorescence boundary line of laser-induced fluorescence emitted by transparent ice. The method involves brightness segmentation of the image, dividing it into brighter and darker regions. Based on the laser's position, the boundary region between the bright and dark areas near the laser is selected as the coarse localization region for the fluorescence boundary line. Within this coarse localization region, the image gradient is calculated, and the direction y of the bright-dark boundary line is calculated based on the gradient. Based on the direction y, an m*d sampling area is set, and within this sampling area, the image boundary E is calculated. This method effectively extracts the coordinates of the laser-induced fluorescence boundary line. Based on the system calibration results, the contour line of the transparent ice body can be measured. Combined with a 3D scanning device, the 3D morphology of the ice can be measured through 3D point cloud stitching, solving the problem that existing line structured light extraction methods cannot be used for 3D morphology measurement of laser-induced fluorescence transparent ice bodies.

Description

Technical Field

[0001] This invention relates to the field of visual measurement technology, and in particular to a method for extracting the fluorescent boundary line of a transparent ice body that emits fluorescence induced by laser. Background Technology

[0002] Aircraft icing and its protection have always been important research topics in the aviation field. To explore icing mechanisms, conduct aerodynamic performance evaluation, safety assessment, and anti-icing / de-icing studies of aircraft under icing meteorological conditions, scholars both domestically and internationally have carried out extensive research in three areas: CFD numerical calculations, wind tunnel testing, and flight testing. [1-4] Due to the low safety and high cost of flight testing under natural icing conditions, simulated icing tests are currently mainly conducted in icing wind tunnels to assess the performance and safety of aircraft under icing conditions, and to verify the performance of anti-icing / de-icing systems and CFD numerical calculation results. Wind tunnel icing tests typically require measuring information such as the thickness and 3D shape of the icing body. Studies have shown that icing is closely related to parameters such as liquid water content, average droplet diameter, temperature, icing time, flight speed, and angle of attack. [5] To further explore the impact of these parameters on the ice growth process, it is also necessary to conduct online measurements of the 3D ice shape during the ice growth process in the icing wind tunnel test.

[0003] Ice formation types include transparent clear ice, translucent mixed ice, and opaque frost ice. To perform online 3D measurement of ice shape during the ice formation process, it is essential to measure the shapes of these three ice types. [6] Based on whether or not contact is made with the ice, existing ice shape measurement methods can be divided into two categories: contact measurement and non-contact measurement. Contact measurement includes the hot knife method and the modeling method; non-contact measurement includes photogrammetry. [7] Light knife technique [8-12] 3D scanner measurement method [13-16] To achieve online measurement of ice shape in icing wind tunnels, non-contact measurement methods must be employed. Existing non-contact methods mostly acquire natural or optically modulated images of the ice surface for three-dimensional topography measurement. The quality of the imaging image directly affects the ice shape measurement results. Experiments show... [11,12] When using the optical scalpel method for ice shape measurement, the imaging quality of clear ice and mixed ice is poor, making it difficult to achieve three-dimensional ice shape measurement. When imaging the surface of frost ice, mixed ice, and clear ice using line structures, as transparency increases, the light transmittance of the line structure increases, and the imaging quality continuously degrades, severely affecting the accuracy of line structure light position extraction.

[0004] Existing optical ice shape measurement methods include: photogrammetry, optical knife method, and 3D scanner measurement. Among them, photogrammetry... [7]A binocular stereo vision measurement system was constructed using two cameras to perform binocular stereo vision three-dimensional reconstruction of the ice body and complete the three-dimensional ice shape measurement. Binocular stereo vision three-dimensional reconstruction must solve the problem of matching the same pixel pairs in the images captured by the two cameras. When the object being photographed is transparent or semi-transparent, it will lead to image quality degradation and cause binocular stereo vision matching to fail. Therefore, the reference [7] only realized the three-dimensional measurement of frost ice and could not realize the three-dimensional reconstruction of transparent ice bodies.

[0005] NASA began researching the use of laser sheet optical simulation of a thermal knife to measure the cross-sectional profile of ice in 1993. [8,9] A laser sheet is projected onto the surface of an ice body. The intersection of the sheet light and the ice body produces deformed laser streaks, which reflect the shape of the ice body at that cross-section. Images of the laser streaks are captured by a camera, and the coordinates of the centerline of the laser streaks are extracted. The shape of the laser streaks is then calculated based on the geometric relationship between the laser plane and the camera, thus obtaining the cross-sectional profile of the ice body at that location. In China, Zhang Long et al.

[10] A binocular structured light method was used to measure the cross-section of frost and ice. (Wang Bin et al.)

[11] The cross-sectional profile of ice was measured using the optical scalpel method.

[0006] The above studies show that the laser beam method can well meet the requirements for measuring the shape of frost and ice. However, for clear ice and mixed ice with transparent or specular reflective areas, it is difficult for the camera to capture clear laser beams on the ice surface, making it difficult to achieve cross-sectional measurement and three-dimensional shape scanning. Figure 4 As shown, because mixed ice has a certain degree of transparency, when the camera images the image, the transmitted light illuminates the area adjacent to the laser streak, weakening the brightness difference between the laser streak and the adjacent area. This results in unclear streak peaks and difficulty in accurately extracting the coordinates of the streak's center line. Furthermore, during the icing experiment, a water film may exist on the ice surface, causing the ice to become a specular reflective surface at certain angles, also preventing the camera from capturing the laser streak. To address this problem, Kang Hanyu et al.

[12] The BM3D noise reduction method improved the positioning accuracy of the light stripe centerline, but it could not fundamentally eliminate the impact of the degradation of the imaging quality of transparent ice bodies.

[0007] In recent years, 3D scanners have been applied to the 3D shape scanning of ice bodies. [13-15]This technology is used to digitize icy bodies in wind tunnel tests in 3D, for CFD calculation result verification, and for aircraft aerodynamic evaluation. Commercial 3D scanners typically use point, line, and surface structured light for 3D measurement. The principle of 3D measurement based on line structured light is the same as that of laser cutting, the difference being the addition of a 3D scanning device to acquire multiple measurement results and synthesize the 3D shape. Compared to laser cutting, 3D scanners provide denser measurement data, enabling the digitization of the 3D shape of the object being measured. Similarly, existing commercial 3D scanners, which use surface reflection imaging (here referred to as "laser reflection imaging" to distinguish it from the "laser-excited radiation imaging" method described later), can only measure diffuse reflection surfaces and cannot measure specular reflection or transparent objects. Therefore, it is necessary to spray a developer on the ice surface to make it an opaque Lambertian surface before using a 3D scanner to scan the icy shape.

[0008] In summary, existing optical ice shape measurement methods suffer from interference from specular reflection of water films on the ice surface and transmission of light from semi-transparent or transparent ice, resulting in cameras being unable to capture clear images of the ice and hindering three-dimensional measurement of transparent ice. For the optical scalpel method, existing methods employ reflected light imaging; however, specular reflection and transmission degrade the quality of line-structured light images during acquisition, making it difficult to accurately extract the position of the line-structured light beam and thus hindering the measurement of mixed and clear ice shapes.

[0009] To address the challenge of structured light imaging of transparent ice bodies, the principle of laser-induced fluorescence is employed, such as... Figure 1 As shown, an ultraviolet laser is used as the induction light source to project sheet light onto an ice body. The sheet light undergoes an electrochemical reaction with water molecules within the ice, generating fluorescence that radiates outward. The fluorescence is captured by a camera, and using the principle of line structured light measurement, the three-dimensional morphology of the transparent ice body can be measured. However, experiments revealed that the laser-induced fluorescence image on the ice surface differs from traditional line structured light imaging images. Existing line structured light centerline extraction methods cannot be used to extract the line structured light centerline, thus hindering three-dimensional scanning. This is because ice is transparent to ultraviolet and visible light. When the laser sheet light irradiates the ice surface, it penetrates into the ice body, simultaneously reacting with water molecules on the surface and inside, generating fluorescence. This fluorescence is emitted not only on the ice surface but also transmitted through the ice body, resulting in a line structured light image of the ice surface that is not a single-peak distribution, but rather... Figure 2 As shown, the boundary between light and dark areas. Therefore, this invention provides a method for extracting the center of structured light oriented laser-induced fluorescence lines.

[0010] [1] Lin Guiping, Bu Xueqin, Shen Xiaobin. Aircraft icing and anti-icing technology [M]. Beijing: Beijing University of Aeronautics and Astronautics Press, 2016.

[0011] [2] Lee S, Broeren AP, Addy Jr HE, et al. Development of 3D IceAccretion Measurement Method [J]. AIAA Paper, 2012, 2938: 2012.

[0012] [3] Miller D, Potapczuk M, Langhals T. Preliminary Investigation ofIce Shape Sensitivity to Parameter Variations [C] / / 43rd AIAA AerospaceSciences Meeting and Exhibit. American Institute of Aeronautics andAstronautics Reston, Virigina, 2005: 2005-0073.

[0013] [4] Pan Huan, Ai Jianliang. Modeling and simulation of aircraft icing shape prediction [J]. Journal of System Simulation, 2014, 26(1):221-224.

[0014] [5] Campbell SE, Broeren AP, Bragg M B. Sensitivity of aircraftperformance to icing parameter variations [J]. Journal of Aircraft, 2007, 44(5): 1758-1760.

[0015] [6] Yi Xian, Wang Bin, Li Weibin, et al. Research progress on measurement methods of aircraft icing shape [J]. Acta Aeronautica Sinica. 2017, 38(2):13-24.

[0016] [7] Collier P, Dixon L, Fontana D, et al. The use of close rangephotogrammetry for studying ice accretion on aerofoil sections[J].Photogrammetric Record, 1999, 16(94): 671-84.

[0017] [8] Mercer CR, Vargas M, Oldenburg J R. A preliminary study on iceshape tracing with a laser light sheet [J]. Nasa Sti / recon Technical ReportN, 1993, 94.

[0018] [9] Hovenac EA, Vargas M. A laser-based ice shape profilometer for use in icing wind tunnels [J]. Nasa Sti / recon Technical Report N, 1995, 95.

[0019]

[10] Long Z, Longde G, Jianjun Y. Investigation of Ice ShapeMeasurement Technique Based on Laser Sheet and Machine Vision in Icing WindTunnel[C] / / Proceedings of the 2009 Fifth International Conference on Imageand Graphics. IEEE Computer Society, 2009:790-795.

[0020]

[11] Wang Bin, Liu Guihua, Zhang Liping, et al. Ice profile measurement based on line structured light [J]. Experimental Fluid Mechanics, 2016, 30(3):14-20.

[0021]

[12] Kang Hanyu, Liu Guihua, Wang Bin, et al. A rapid method for extracting the center line of laser light band on ice-shaped surfaces [J]. Experimental Fluid Mechanics, 2017, 31(5): 81-87.

[0022]

[13] Lee S, Broeren AP, Addy Jr HE, et al. Development of 3D IceAccretion Measurement Method [J]. AIAA Paper, 2012, 2938: 2012. Summary of the Invention

[0023] The purpose of this invention is to address the shortcomings of existing technologies by providing a method for extracting the fluorescence boundary line of laser-induced transparent ice bodies. This method involves brightness segmentation of the image, dividing it into brighter and darker regions. Based on the laser's position, the boundary region between the bright and dark areas near the laser is selected as the coarse localization region for the fluorescence boundary line. Within this coarse localization region, the image gradient is calculated, and the direction y of the bright-dark boundary line is calculated based on the gradient. Based on the direction y, an m*d sampling area is set, and within this sampling area, the image boundary E is calculated. This method effectively extracts the coordinates of the laser-induced fluorescence boundary line. Based on the system calibration results, the contour line of the transparent ice body can be measured. Combined with a 3D scanning device, the 3D morphology of the ice body can be measured through 3D point cloud stitching, thus solving the problem that existing line structured light extraction methods cannot be used for 3D morphology measurement of laser-induced fluorescence transparent ice bodies.

[0024] This solution is achieved through the following technical measures:

[0025] A method for extracting the fluorescent boundary line of a laser-induced transparent ice body emitting fluorescence includes the following steps:

[0026] a. Perform brightness segmentation on the captured laser-induced fluorescence image of the ice body, dividing the image into two regions: a brighter region and a darker region;

[0027] b. Based on the location of the laser, select the bright and dark boundary area near the laser as the coarse positioning area for the fluorescence boundary line;

[0028] c. Based on the coarse localization area, take one pixel, calculate the image gradient at that pixel, and calculate the direction y of the fluorescence boundary line based on the gradient;

[0029] d. Along the direction y, set a rectangular sampling area of ​​m*d, where m is the width and d is the height. The width is perpendicular to the direction y of the fluorescence boundary line. Use a gradient-based edge detection method to find the image edge coordinates. After traversing all pixels in the coarse positioning area, obtain the fine positioning coordinates, which are the fluorescence boundary line coordinates.

[0030] As a preferred embodiment of this scheme: during laser-induced fluorescence imaging, two images are captured simultaneously: a reference image and a measurement image; when capturing the reference image, the laser is turned off to capture the substrate signal; when capturing the measurement image, the laser is turned on to capture the laser-induced fluorescence image; by comparing the two images, the laser-induced fluorescence imaging area can be found, which facilitates the localization of the laser-induced fluorescence area.

[0031] As a preferred approach to this solution: before performing step a, the image is subjected to median filtering to eliminate image noise interference and improve the image signal-to-noise ratio.

[0032] As a preferred embodiment of this scheme: In step c, before performing gradient calculation, an image similarity matching method is used to find similar image blocks within the coarse localization area of ​​the fluorescence boundary line. Then, a weight matrix is ​​constructed based on the pixel values ​​and distances within the image blocks to denoise the current image, thereby enhancing the signal-to-noise ratio of the fluorescence boundary line. After the signal-to-noise ratio of the fluorescence boundary line is enhanced, subsequent calculations are performed.

[0033] As a preferred approach to this scheme: when performing image denoising, SVD decomposition is performed on the found similar image blocks to find the top k principal components, and the image is reconstructed to eliminate image noise interference.

[0034] As a preferred option in this solution, reconstructing the light bar region of the captured image can further improve the scope and accuracy of image patch similarity matching search.

[0035] As a preferred approach to this solution, when performing image denoising, the BM3D method is used to denoise the similar image blocks found.

[0036] As a preferred approach to this solution: when performing image denoising, similar image patch matching is performed not only in the current image, but also in all acquired images within the coarse localization region. Based on a large number of similar image patches, line structured light image reconstruction is performed.

[0037] As a preferred approach to this scheme, when performing image denoising, the LLC method is used to denoise the similar image blocks found.

[0038] As a preferred approach to this scheme: when performing similar image block search and matching, instead of determining similarity based on the image's own coordinate system, the total direction y0 of the image block to be denoised is calculated along the coarse positioning line trajectory, and a sampling area S0 with the same direction as y0 is set; along the coarse positioning line trajectory, a sampling area Si is set according to the direction of the coarse positioning line, and image blocks are extracted for similarity comparison.

[0039] The beneficial effects of this solution can be understood from the description of the solution above:

[0040] By calculating the image gradient in the coarse localization region and calculating the direction yj of the boundary line based on the gradient, the sampling region is set, and finally, the coordinates of the fluorescence boundary line are finely localized.

[0041] By simultaneously acquiring reference and measurement images, the problem of incomplete and erroneous extraction of laser-induced fluorescence regions caused by inhomogeneities within the transparent ice body is reduced, thus improving the accuracy of coarse localization of the fluorescence boundary lines of laser-induced fluorescence.

[0042] A fine-line fluorescence boundary line localization image denoising method is proposed, including image median filtering and NLM, BM3D, and GPCA algorithms based on line structured light for coarse localization regions. These algorithms further improve the signal-to-noise ratio of laser-induced fluorescence boundaries, thereby further improving the extraction accuracy of line fluorescence boundaries.

[0043] Meanwhile, these algorithms differ from their existing usage when applied: instead of searching for similar image patches in the entire image, this invention only performs search and matching within the coarse localization area, reducing the search space and obtaining image patches with similar physical meanings, thus achieving better noise reduction results compared to existing image patch search methods that search the entire image.

[0044] Therefore, it is evident that the present invention has substantial features and progress compared with the prior art, and the beneficial effects of its implementation are also obvious. Attached Figure Description

[0045] Figure 1 This is a schematic diagram illustrating the principle of laser-induced fluorescence ice shape measurement.

[0046] Figure 2 A schematic diagram of a camera capturing fluorescence in ice.

[0047] Figure 3 This is a schematic diagram of the boundary between fluorescence and the surface of ice.

[0048] Figure 4 This is a schematic diagram of the sampling area set on the coarse positioning area in this invention;

[0049] Figure 5 A schematic diagram showing the coordinates for locating the fluorescence boundary line obtained in this invention;

[0050] Figure 6 This is a schematic diagram illustrating the search for image blocks similar to the image block to be denoised in Example 4;

[0051] Figure 7 This is a schematic diagram of similar image patch search and matching in Example 5.

[0052] In the figure, 1 is the camera, 2 is the filter, 3 is the laser, 4 is the ice body, 5 is the sheet light, 6 is the laser-induced fluorescence, 8 is the laser-induced fluorescence interface, 9 is the fluorescence boundary line, 10 is the non-laser-induced region, 11 is the ice body cross section, 12 is the coarse positioning region, 13 is the sampling region, 14 is the image block to be denoised, 15 is the image block to be matched, and 16 is the fine positioning coordinates. Detailed Implementation

[0053] All features disclosed in this specification, or steps in all methods or processes disclosed herein, may be combined in any way, except for mutually exclusive features and / or steps.

[0054] Any feature disclosed in this specification (including any appended claims, abstract, and drawings) may be replaced by other equivalent or similar features for a similar purpose, unless specifically stated otherwise. That is, unless specifically stated otherwise, each feature is merely one example of a series of equivalent or similar features.

[0055] Example 1

[0056] Combination Figures 3-5 The operation steps of this method are as follows:

[0057] a. Perform brightness segmentation on the captured laser-induced fluorescence image of the ice body, dividing the image into two regions: a brighter region and a darker region;

[0058] b. Based on the location of the laser, select the bright and dark boundary area near the laser as the coarse positioning area for the fluorescence boundary line;

[0059] c. Based on the coarse localization area, take one pixel, calculate the image gradient at that pixel, and calculate the direction y of the fluorescence boundary line based on the gradient;

[0060] d. Along the direction y, set a rectangular sampling area of ​​m*d, where m is the width and d is the height. The width is perpendicular to the direction y of the fluorescence boundary line. Use a gradient-based edge detection method to find the image edge coordinates. After traversing all pixels in the coarse positioning area, obtain the fine positioning coordinates, which are the fluorescence boundary line coordinates.

[0061] Example 2

[0062] The difference from Example 1 is that two images are captured simultaneously during laser-induced fluorescence imaging: a reference image and a measurement image. When capturing the reference image, the laser is off, and the substrate signal is captured; when capturing the measurement image, the laser is on, and the laser-induced fluorescence image is captured. By comparing the two images, the laser-induced fluorescence imaging region can be located, facilitating the localization of the laser-induced fluorescence region. Based on the localization results, the fluorescence boundary line is further calculated.

[0063] Example 3

[0064] The difference from Example 1 is that, before performing the first step, the image is subjected to median filtering to eliminate image noise interference and improve the image signal-to-noise ratio.

[0065] Example 4

[0066] The difference from Example 1 is that in step three, within the coarse localization region along the structured light line, image blocks similar to the current image block are searched along the coarse localization line trajectory to form a set of similar image blocks. The weight matrix in the NLM is sampled to denoise the image block to be denoised, thereby enhancing the signal-to-noise ratio of the fluorescence boundary line. After enhancement, subsequent calculations are performed.

[0067] like Figure 6 As shown, similar image patches are found along the coarse positioning line. The criterion for determining similar image patches is that the Euclidean distance is less than a certain set threshold.

[0068] Example 5

[0069] The difference from Example 4 is that, when performing similar image block search and matching, the similarity judgment is not based on the image's own coordinate system. Instead, after calculating the total direction y0 of the "to be denoised image block" along the coarse positioning line trajectory, a sampling area S0 with the same direction as y0 is set. Similarly, in the coarse positioning line trajectory, the sampling area is set according to the direction of the coarse positioning line, and the image blocks are extracted for similarity comparison.

[0070] pass Figure 7 As can be seen from Example 4, the sampling area Si is aligned along the fluorescence boundary line, which has stronger similarity and can find more similar image blocks, thus achieving better image boundary denoising effect.

[0071] Example 6

[0072] The difference from Example 4 is that, when performing image denoising, SVD decomposition is performed on the found similar image blocks to find the top k principal components, and the image is reconstructed to eliminate image noise interference.

[0073] Example 7

[0074] Replace the SVD decomposition in Example 6 with the NLM algorithm.

[0075] Example 8

[0076] Replace the SVD decomposition in Example 6 with the BM3D algorithm.

[0077] Example 9

[0078] Replace the SVD decomposition in Example 6 with the LLC algorithm.

[0079] Example 10

[0080] The difference from Example 4 is that the light stripe area of ​​the images captured throughout the experiment is used for reconstruction, which further improves the range and accuracy of image block similarity matching search.

[0081] This invention is not limited to the specific embodiments described above. The invention extends to any new feature or combination disclosed in this specification, as well as any new method or process step or combination disclosed herein.

Claims

1. A method for extracting the fluorescent boundary line of a laser-induced transparent ice body emitting fluorescence, characterized in that: It includes the following steps: a. Perform brightness segmentation on the captured laser-induced fluorescence image of the ice body, dividing the image into two regions: a brighter region and a darker region; b. Based on the location of the laser, select the bright and dark boundary area near the laser as the coarse positioning area for the fluorescence boundary line; c. Based on the coarse localization area, take one pixel, calculate the image gradient at that pixel, and calculate the direction y of the fluorescence boundary line based on the gradient; d. Along the direction y, set a rectangular sampling area of ​​m*d, where m is the width and d is the height. The width is perpendicular to the direction y of the fluorescence boundary line. Use a gradient-based edge detection method to find the image edge coordinates. After traversing all pixels in the coarse localization area, obtain the fine localization coordinates, which are the fluorescence boundary line coordinates. During laser-induced fluorescence imaging, two images are captured simultaneously: a reference image and a measurement image. When capturing the reference image, the laser is turned off to capture the substrate signal. When capturing the measurement image, the laser is turned on to capture the laser-induced fluorescence image. By comparing the two images, the laser-induced fluorescence imaging area can be located, which facilitates the localization of the laser-induced fluorescence area. In step c, before gradient calculation, image similarity matching is used to find similar image blocks within the coarse localization area of ​​the fluorescence boundary line. Then, a weight matrix is ​​constructed based on the pixel values ​​and distances within the image blocks to denoise the current image, thereby enhancing the signal-to-noise ratio of the fluorescence boundary line. After the signal-to-noise ratio of the fluorescence boundary line is enhanced, subsequent calculations are performed.

2. The method for extracting the fluorescence boundary line of a laser-induced transparent ice body emitting fluorescence according to claim 1, characterized in that: Before performing step a, the image is subjected to median filtering to eliminate image noise interference and improve the image signal-to-noise ratio.

3. The method for extracting the fluorescent boundary line of a laser-induced transparent ice body emitting fluorescence according to claim 1, characterized in that: When performing image denoising, SVD decomposition is performed on the found similar image blocks to find the top k principal components, and the image is reconstructed to eliminate image noise interference.

4. The method for extracting the fluorescent boundary line of a laser-induced transparent ice body emitting fluorescence according to claim 1, characterized in that: Reconstructing the light bar region of the captured image can further expand the scope and accuracy of image patch similarity matching search.

5. The method for extracting the fluorescent boundary line of a laser-induced transparent ice body emitting fluorescence according to claim 1, characterized in that: When performing image denoising, the BM3D method is used to denoise similar image patches that are found.

6. The method for extracting the fluorescence boundary line of a laser-induced transparent ice body emitting fluorescence according to claim 1, characterized in that: When performing image denoising, similar image patch matching is performed not only in the current image, but also in all acquired images within the coarse localization region. Based on a large number of similar image patches, line structured light image reconstruction is performed.

7. The method for extracting the fluorescent boundary line of a laser-induced transparent ice body emitting fluorescence according to claim 1, characterized in that: When performing image denoising, the LLC method is used to denoise similar image patches that are found.

8. The method for extracting the fluorescent boundary line of a laser-induced transparent ice body emitting fluorescence according to claim 1, characterized in that: in When performing similar image block search and matching, the similarity judgment is not based on the image's own coordinate system. Instead, the total direction y0 of the image block to be denoised is calculated along the coarse positioning line trajectory, and a sampling area S0 with the same direction as y0 is set. Along the coarse positioning line trajectory, a sampling area Si is set according to the direction of the coarse positioning line, and the image blocks are extracted for similarity comparison.

Images