An automatic resolution discrimination method based on USAF 1951 resolution target

Abstract

The application discloses a kind of resolution automatic discrimination methods based on USAF1951 resolution target, belong to image processing technical field, the algorithm includes image positioning segmentation step and resolution discrimination step, image positioning segmentation step is positioned by local Gaussian threshold method and edge detection to determine the square block part in square block target, then the angle of image is corrected and segmented to extract stripe cell image, resolution discrimination step is compared with standard stripe image after conversion to frequency domain with segmented stripe image, determines the similarity of each stripe cell, and obtains curve by polynomial fitting sequence, finally outputs the resolution of night vision device.Compared with prior art, the application can improve the use efficiency while meeting the accuracy requirements.

Description

Technical Field

[0001] This invention belongs to the field of image processing technology, and in particular to an automatic, efficient, objective and accurate method for determining the resolution of a target based on USAF1951 resolution. Background Technology

[0002] Resolution is related to the spatial frequency of the modulation transfer function curve of an optical device. It reflects the working distance and image sharpness of the optoelectronic device under different illumination conditions and is one of the important parameters for judging the performance of an imaging system. The objective resolution evaluation technique based on the USAF 1951 resolution target is a technique used to detect the resolution of an imaging system and is widely used in optical engineering, image measurement and other fields.

[0003] The primary method for measuring resolution remains subjective evaluation. Professional testers observe the USAF-1951 resolution target image displayed on the night vision device's output and identify the smallest identifiable stripe unit in the resolution image. The spatial resolution corresponding to this stripe group is an approximate estimate of the low-light night vision device's resolution. This method requires multiple experiments by professional testers to reach an effective judgment. Since subjective evaluation is related to the tester's vision, psychology, and experience, objective evaluation methods have been continuously researched. Liu Zhengyun et al. used gradient entropy to objectively evaluate the texture features of stripe images; Ai Juan et al. proposed an objective evaluation method based on Fast Fourier Transform, converting the contrast of the stripe image into the first harmonic amplitude of the spectral image for quantification; Shi Jifang et al. combined a normalized cross-correlation model and an optical modulation model to estimate the resolution of low-light image intensifiers.

[0004] While these objective resolution evaluation algorithms can provide objective assessments of target stripes, several issues remain. First, existing methods still rely heavily on manual intervention, often requiring manual selection of the stripes or target image. This is not only inefficient, but even minor deviations in the cutting position can significantly impact the test results. Second, accuracy and time consumption remain challenges, hindering practical application. Therefore, reducing reliance on manual intervention and further improving the accuracy and efficiency of resolution assessment imaging systems is a crucial technical challenge in this field. Summary of the Invention

[0005] This invention provides an automatic resolution discrimination method based on USAF1951 resolution targets to improve the accuracy and efficiency of resolution discrimination.

[0006] The technical solution to achieve the purpose of this invention is as follows:

[0007] An automatic resolution discrimination method based on USAF 1951 resolution targets, comprising the following steps:

[0008] Step 1: Automatically crop the acquired image. First, binarize the original image, then extract the contour to obtain the largest inscribed rectangle of the circular contour in the original image. After obtaining the region of interest, use Gaussian filtering for further noise reduction.

[0009] Step 2: Locate the block in the USAF1951 resolution target, solve the angular coordinates of the block, and mark the outline of the block in the filtered inscribed rectangle diagram.

[0010] Step 3: Perform adaptive rotation on the block outline positioning image to obtain the adaptively rotated block outline positioning image;

[0011] Step 4: Extract individual unit stripe images from the adaptively rotated block contour localization map;

[0012] Step 5: Transform the unit stripe image and the stripe image in the USAF1951 resolution target of the same scale into the frequency domain through Fourier transform, and calculate the similarity of the stripe image spectrum.

[0013] Step 6: Establish a similarity sequence for all stripe units in the unit stripe image, perform polynomial curve fitting using the least squares method, and calculate the resolution using a threshold.

[0014] Compared with the prior art, the significant advantages of this invention are:

[0015] (1) Automatic and efficient: The resolution target is automatically positioned and segmented to obtain each stripe unit. The automation process reduces the reliance on manual labor and improves efficiency and repeatability.

[0016] (2) High resolution and discrimination accuracy: The identifiability of the unit stripe pattern is evaluated by using the spectral similarity between the stripe to be tested and the standard stripe, and the resolution of the imaging system to be tested is obtained by polynomial fitting. Attached Figure Description

[0017] Figure 1 To use a night vision device to measure the light intensity Actual image acquired by a target plate with a contrast ratio of 0.35 at lx.

[0018] Figure 2 This is the image after binarization of the acquired image.

[0019] Figure 3 (a) is the largest inscribed rectangle. Figure 3 (b) is the inscribed rectangle after filtering.

[0020] Figure 4 The image is processed using the local Gaussian thresholding method.

[0021] Figure 5The images are morphologically filtered, where (a) is the image after erosion and (b) is the image after dilation.

[0022] Figure 6 (a) is the image after Canny edge detection. Figure 6 (b) is the image after re-dilation.

[0023] Figure 7 This is a location map of the block outline.

[0024] Figure 8 This is the positioning map of the cube outline after adaptive rotation.

[0025] Figure 9 This is a segmentation result of the striped image.

[0026] Figure 10 Fitting curves for the stripe similarity of target plates with different illuminance and contrast under the vehicle-length observation lens.

[0027] Figure 11 This is a flowchart of the present invention. Detailed Implementation

[0028] In this embodiment, an automatic resolution discrimination method based on a USAF1951 resolution target includes the following steps:

[0029] Step 1: Automatically crop the acquired image. First, binarize the original image, then extract the contour to obtain the largest inscribed rectangle of the circular contour in the original image. After obtaining the region of interest, Gaussian filtering is used for further noise reduction.

[0030] The process involves placing the camera at the eyepiece of the imaging system under test, observing the USAF1951 target in the measuring instrument, and continuously acquiring ten frames of images, which are then averaged and used as the original image. Specifically, the automatic cropping process consists of the following steps:

[0031] Step 1.1: Binarize the original image;

[0032] The original image collected is as follows Figure 1 As shown, the binarized image is as follows Figure 2 As shown.

[0033] Step 1.2: Extract the contours from the binarized image to obtain the maximum inscribed rectangle of the circular contour in the original image;

[0034] The maximum inscribed rectangle of the circular outline is shown below. Figure 3 As shown, the pixels of the inscribed rectangle are reset to (640×640).

[0035] Step 1.3: Apply Gaussian filtering to the maximum inscribed rectangle to obtain the filtered inscribed rectangle;

[0036] Step 2: Locate the cube in the USAF1951 resolution target, solve for the cube's angular coordinates, and mark the cube's outline in the filtered inscribed rectangle diagram.

[0037] Specifically, the block positioning involves the following steps:

[0038] Step 2.1: Use the local Gaussian thresholding method to perform threshold segmentation on the filtered inscribed rectangle to separate the background and foreground parts;

[0039] The pixels in the inscribed rectangle after filtering Centered on the nearest point, the width and height of the neighborhood window are respectively and Within a given region, the local mean and standard deviation can be expressed as:

[0040]

[0041]

[0042] In the formula, It is a local mean. Standard deviation, Let i be the coordinates of pixel I, i be the magnitude of the horizontal movement of pixels within the neighborhood, and j be the magnitude of the vertical movement of pixels within the neighborhood.

[0043] The local threshold can be expressed as:

[0044]

[0045] in, For local thresholds, These are custom parameters used to adjust the sensitivity of the threshold. Generally, The value ranges from 0.5 to 2.

[0046] Compare the grayscale value of the pixel with the local threshold. A comparison is made to distinguish the foreground from the background. The image is processed using the local Gaussian thresholding method, as shown below. Figure 4 As shown.

[0047] Step 2.2: Apply morphological filtering to the image processed by the local Gaussian thresholding method to correct the interior and boundaries of the blocks in the image processed by the local Gaussian thresholding method.

[0048] Because the local adaptive thresholding method is related to the average pixel value of a certain surrounding area, it can easily cause central holes in the image. Therefore, it is necessary to fill the gaps between edges and fill in incomplete contours. An opening operation is performed on the image processed by the local Gaussian thresholding method to obtain the eroded image, as shown below. Figure 5As shown in (a), further dilation is performed to obtain the corrected positioning map as follows. Figure 5 As shown in (b).

[0049] Step 2.3: Use Canny edge detection to obtain the edges in the image from the corrected localization map. After obtaining the edges, dilate them to close the contours, and then select the contours with the largest areas to obtain the precise contours of the square.

[0050] The image after Canny edge detection is as follows: Figure 6 As shown in (a), the redilated image is as follows Figure 6 As shown in (b).

[0051] Step 2.4: Perform a Hough transform on the precise outline of the block to detect straight lines and calculate intersection points, obtaining the corner coordinates of the block. Finally, mark the block outline in the filtered inscribed rectangle diagram: the block outline positioning diagram is shown below. Figure 7 As shown.

[0052] Step 3: Perform adaptive rotation on the block outline positioning image to obtain the adaptively rotated block outline positioning image.

[0053] After locating the block, since the corner coordinates of the block have been obtained, the rotation angle can be calculated using the coordinates. Finally, the image is rotated with the center point of the block's outline positioning image as the center point.

[0054] The rotated portion of the image needs to be cropped, and the final pixel resolution of the rotated image is reset to (640×640). Additionally, some non-integer points are calculated during the mapping coordinate calculation, necessitating interpolation to determine their pixel values. This patent employs bilinear interpolation to achieve the image rotation process.

[0055] Rotation requires two coordinate transformations: first, the image coordinate system is converted to a Cartesian coordinate system, and then the rotation... After determining the angle, calculate the coordinates of the specified pixel after rotation, and finally convert it back to the image coordinate system. The conversion formula is shown below.

[0056]

[0057]

[0058]

[0059] in, and These represent the x and y coordinates of any pixel in the original image. and These represent the x and y coordinates of the pixels in the rotated image. Indicates the angle to be rotated. and These are the width and height of the block outline positioning map, respectively. and These represent the width and height of the adaptively rotated block outline positioning image, respectively. The adaptively rotated block outline positioning image is shown below. Figure 8 As shown.

[0060] Step 4: Extract individual unit stripe images from the adaptively rotated block contour localization map.

[0061] Based on the proportional relationships of each element in the USAF 1951 resolution target, independent element fringe images are extracted from the adaptively rotated block contour localization image (the block target under test). After the initial processing steps, the positions of the element fringe in the block contour localization image are basically the same as the fringe positions in the USAF 1951 resolution target. When extracting and cropping the block contour localization image, the cropping unit should be slightly larger than the element fringe image to ensure that all fringe elements are within the cropping unit.

[0062] Based on the distribution of elements in the USAF 1951 resolution target, assuming the side length of the square is *a* (pixels) and the center coordinates of the square are... Based on the positional relationships of each group in the USAF 1951 resolution target, the coordinates of the upper left corner of the stripe images in groups 1-1 to 3-6 are as follows: , , , , , , , , , , , . , , , , , .

[0063] Based on the structural relationship between the stripes, the segmentation result of the stripe image is as follows: Figure 9 As shown.

[0064] Step 5: Transform the unit stripe image and the stripe image in the USAF1951 resolution target of the same scale into the frequency domain through Fourier transform, and calculate the similarity of the stripe image spectrum.

[0065] After extracting the unit fringe image, fringe images of different groups were obtained. The unit fringe images of different groups and the standard fringe images of the same scale were then transformed to the frequency domain using Fourier transform.

[0066]

[0067]

[0068] It corresponds to spatial coordinates The frequency domain representation, This represents a stripe image in a two-dimensional spatial domain of size M×N, where , and , and These correspond to the frequency components of the image in the horizontal and vertical directions. In the two-dimensional discrete Fourier transform, each pair... Each corresponds to a specific frequency component. The result of the transformation is a complex matrix, where each element... Each has a real part and an imaginary part These represent the amplitudes of the cosine (real part) and sine (imaginary part) waves of that frequency component, respectively. Modulus Represents frequency The amplitude of the amplitude is the Fourier spectrum.

[0069] Choosing to calculate the Euclidean distance between spectrograms as the method for determining similarity, spectrogram similarity can be expressed as:

[0070]

[0071] In the above formula This refers to the similarity of the spectrograms. The spectral amplitude of a target stripe image at standard USAF 1951 resolution. The amplitude of the spectrum of the unit stripe image.

[0072] Step 6: Establish a similarity sequence for all stripe units in the unit stripe image, perform polynomial curve fitting using the least squares method, and calculate the resolution using a threshold.

[0073] Step 6.1: Calculate the spectral similarity of all stripe units in the unit stripe image, and use the one-to-one correspondence between stripe units and spectral similarity values ​​as the similarity sequence; that is, each stripe unit corresponds to a spectral similarity value.

[0074] Step 6.2: Use the least squares method to fit the similarity sequence to a polynomial curve and calculate the resolution by using a threshold.

[0075] Since higher-level stripe groups have a greater impact from noise, the similarity fluctuates more in high-resolution stripe groups, and may even decrease drastically. Therefore, stripe groups with a similarity of 0.2 or higher in the similarity sequence are selected, and those that do not conform to the decreasing trend of each group are removed. Subsequently, the available similarity sequence data are fitted using multiple polynomials to obtain the stripe group similarity fitting curve.

[0076] Let the threshold T = 0.497 (obtained empirically). When the similarity of stripe groups is greater than the threshold, they are considered identifiable; when the similarity is less than the threshold, they are considered unidentifiable. The resolution can be obtained by substituting the threshold T into the stripe group similarity fitting curve function.

[0077] Example

[0078] Taking the WG1013 vehicle length observation mirror as an example, the similarity fitting curves of the stripe groups of target plates with different illuminance and contrast under the vehicle length observation mirror are as follows: Figure 10 As shown.

[0079] In light When the lx and target-plate contrast are 0.85, a cubic polynomial curve can be obtained by polynomial fitting based on the similarity numerical sequence: ; under light conditions When lx and target plate contrast are 0.35, the analytical expression of the fitting curve is: Substituting the threshold value of 0.497, we obtain a resolution of 5.799 lp / mm under high illumination and high contrast target conditions, and a resolution of 3.087 lp / mm under low illumination and low contrast target conditions.

Claims

1. An automatic resolution discrimination method based on a USAF 1951 resolution target, characterized in that the steps include... include: Step 1: Automatically crop the acquired image. First, binarize the original image, then extract the contour to obtain the largest inscribed rectangle of the circular contour in the original image. After obtaining the region of interest, use Gaussian filtering to remove noise. Step 2: Locate the block in the USAF1951 resolution target, solve the angular coordinates of the block, and mark the outline of the block in the filtered inscribed rectangle diagram. Step 3: Perform adaptive rotation on the block outline positioning image to obtain the adaptively rotated block outline positioning image; Step 4: Extract individual unit stripe images from the adaptively rotated block contour localization map; Step 5: Transform the unit stripe image and the stripe image in the USAF1951 resolution target of the same scale into the frequency domain through Fourier transform, and calculate the similarity of the stripe image spectrum. The formula for converting the unit fringe image to be measured and the standard fringe image of the same scale to the frequency domain is: ； ； It corresponds to spatial coordinates The frequency domain representation, This represents a stripe image in a two-dimensional spatial domain of size M×N. and Corresponding to the frequency components of the image in the horizontal and vertical directions; in the two-dimensional discrete Fourier transform, each pair Each corresponds to a specific frequency component; the result of the transformation is a complex matrix, where each element... Each has a real part and an imaginary part These represent the amplitudes of the cosine and sine waves of that frequency component, respectively; the modulus... Represents frequency The amplitude of this amplitude is the Fourier spectrum. The Euclidean distance between spectrograms is chosen as the method for determining similarity. Spectrogram similarity is expressed as follows: ； In the above formula This refers to the similarity of the spectrograms. The spectral amplitude of a target stripe image at standard USAF 1951 resolution. The amplitude of the spectrum of the unit stripe image; Step 6: Establish a similarity sequence for all stripe units in the unit stripe image, perform polynomial curve fitting using the least squares method, and calculate the resolution using a threshold.

2. The automatic resolution discrimination method based on a USAF 1951 resolution target according to claim 1, characterized in that, In step 2, Step 2.1: Use the local Gaussian thresholding method to perform threshold segmentation on the filtered inscribed rectangle to separate the background and foreground parts; Step 2.2: Apply morphological filtering to the image processed by the local Gaussian thresholding method to correct the interior and boundaries of the blocks in the image processed by the local Gaussian thresholding method. Step 2.3: Use Canny edge detection to obtain the edges in the image from the corrected localization map. After obtaining the edges, dilate them to close the contours, and then select the contours with the largest areas to obtain the precise contours of the square. Step 2.4: Perform Hough transform on the precise outline of the block to detect straight lines and calculate the intersection points to obtain the corner coordinates of the block. Finally, mark the block outline in the filtered inscribed rectangle diagram.

3. The automatic resolution discrimination method based on a USAF 1951 resolution target according to claim 1, characterized in that, In step 3, the block outline positioning image is adaptively rotated. This rotation requires two coordinate transformations: first, the image coordinate system is converted to a Cartesian coordinate system, and then the rotation... After determining the angle, calculate the coordinates of the specified pixel after rotation, and then convert it back to the image coordinate system; the two transformation formulas are as follows: ； ； ； in, and These represent the x and y coordinates of any pixel in the original image. and These represent the x and y coordinates of the pixels in the rotated image. Indicates the angle to be rotated. and These are the width and height of the block outline positioning map, respectively. and These are the width and height of the adaptively rotated block outline positioning image, respectively.

4. The automatic resolution discrimination method based on a USAF 1951 resolution target according to claim 1, characterized in that, In step 4, when extracting the individual unit stripe images, the cropping unit should be slightly larger than the unit stripe image when extracting and cropping the block contour positioning map to ensure that the stripes are all within the cropping unit.

5. The automatic resolution discrimination method based on a USAF 1951 resolution target according to claim 1, characterized in that, The specific steps of step 6 are as follows: Step 6.1: Calculate the spectral similarity of all stripe units in the unit stripe image, and use the one-to-one correspondence between stripe units and spectral similarity values ​​as the similarity sequence; Step 6.2: Use the least squares method to fit the similarity sequence to a polynomial curve and calculate the resolution by using a threshold.

6. The automatic resolution discrimination method based on a USAF 1951 resolution target according to claim 4, characterized in that, Based on the distribution of elements in the USAF 1951 resolution target, assuming the side length of the square is pixel *a*, and the center coordinates of the square are... Based on the positional relationships of each group in the USAF 1951 resolution target, the coordinates of the upper left corner of the stripe images in groups 1-1 to 3-6 are as follows: , , , , , , , , , , , ; , , , , , .

7. The automatic resolution discrimination method based on a USAF 1951 resolution target according to claim 5, characterized in that, Let the threshold T = 0.

497. When the similarity of stripe groups is greater than the threshold, they are considered identifiable. When the similarity of stripe groups is less than the threshold, they are considered unidentifiable. The resolution can be obtained by substituting the threshold T into the stripe group similarity fitting curve function.

Images