Infrared simulation image quality evaluation method based on random forest

By using random forest algorithm and multi-angle feature extraction in infrared simulation image quality evaluation, the problem of ignoring visual features and artificial weights in the existing technology is solved, and a more accurate and objective infrared simulation image quality evaluation is achieved, which improves the evaluation. Realism.

CN114782389BActive Publication Date: 2024-11-01XIDIAN UNIV
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
CN202210474340.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-04-29
Publication Date
2024-11-01
Estimated Expiration
2042-04-29

AI Technical Summary

Technical Problem

The prior art ignores human visual characteristics in infrared simulation image quality evaluation, cannot comprehensively summarize the overall characteristics of the image, and the evaluation lacks objectivity due to artificial weighting.

Method used

Using a random forest-based method, the pixel brightness, structure, texture and shape characteristics of infrared simulation images are extracted, and the model is trained through the random forest algorithm, and the weights of each feature are given different according to human subjective cognition to generate more in line with the perception of the human eye. Quality evaluation.

Benefits of technology

A more accurate and objective evaluation of the quality of infrared simulation images is achieved, and the evaluation results are more in line with human visual perception and have higher fidelity.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114782389B_ABST
    Figure CN114782389B_ABST
Patent Text Reader

Abstract

The present invention discloses a method for evaluating the quality of infrared simulation images based on random forests, which mainly solves the problem that the prior art lacks objective evaluation criteria and ignores human visual features, resulting in the inconsistency between the quality evaluation results of infrared simulation images and human subjective perception. The implementation steps of the present invention are: acquiring an infrared simulation image; extracting pixel brightness features, structural features, shape features, and texture features of the infrared simulation image respectively; and using a random forest regressor for training and prediction. The quality evaluation results of infrared simulation images of the present invention are more consistent with the visual perception of the human eye, and the evaluation results are more accurate. The present invention can be widely used in the field of infrared image processing technology.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of image processing technology, and further relates to a method for evaluating the quality of infrared simulation images based on random forests in the field of infrared simulation image processing technology. The present invention can be used to evaluate the quality of infrared simulation images. Background Art

[0002] Infrared simulation image quality evaluation has important applications in image processing, simulation system verification and other fields. At present, although there are many evaluations on the simulation effects of infrared scene simulation images, they mainly evaluate and calibrate each module during the simulation process, or only verify and analyze specific targets or backgrounds according to project requirements. More often, the simulation effects still need to be identified and certified by experts, and there is no objective evaluation standard. At the same time, the evaluation of the quality of simulated images still uses traditional image quality evaluation methods and means, and it is impossible to objectively evaluate the simulated images by multi-feature quantification technical means.

[0003] Yunnan North Optical Technology Co., Ltd. disclosed a method for evaluating the clarity of infrared simulated images based on the human visual system in its patent application "A method for evaluating the clarity of infrared simulated images based on the human visual system without reference" (patent application number: CN202110931781.6, application publication number: CN113570596A). The method first extracts the gradient information of the original image and the reference image, and then divides the gradient image into N blocks. The corresponding weights are obtained by the relative size of the impulses caused in the visual nerves of each block in the gradient image and the human visual system filter; then the structural similarity of the gradient image and the corresponding blocks of the image is calculated; finally, the clarity value of the original image is obtained by the weight and structural similarity. Although this method can be used to evaluate the quality of infrared simulated images, the method still has the disadvantage that it only extracts image features from the two aspects of image gradient and structural similarity, ignoring the characteristic information carried by the radiation brightness and image content in the image, and not considering the different proportions of different features in human vision, which does not conform to the characteristics of human vision. It is impossible to summarize the overall characteristics of infrared simulated images.

[0004] Zhejiang Tongshan Artificial Intelligence Technology Co., Ltd. disclosed a method for judging the image quality that can be used for infrared simulation images in its patent document "An image quality judgment method, device and storage medium" (patent application number: CN202010602500.8, application publication number: CN 111798421 A). The specific steps of implementing this method are: the first step is to evaluate the quality of the image from three aspects: gradient, pixel features, and spectral features; the second step is to manually assign weights to the features extracted from the three aspects of gradient, pixel features, and spectral features; the third step is to establish an image quality comprehensive evaluation model to obtain a comprehensive image quality score. Although this method can be used to evaluate the image quality of infrared simulation images, the method still has the disadvantage that although this method describes the same image from multiple aspects of gradient, pixel features, and spectral features, it uses a manual method to artificially assign weights to the evaluation indicators of each feature. The artificial subjective evaluation leads to the lack of objective judgment criteria for the quality evaluation of infrared simulation images. Summary of the invention

[0005] The purpose of the present invention is to propose an infrared simulation image quality evaluation method based on random forest in view of the above-mentioned deficiencies of the prior art, so as to solve the problem that the prior art ignores the human visual characteristics and cannot summarize the overall characteristics of the infrared simulation image, and artificially assigns weights to the evaluation indicators of each feature, resulting in the lack of objective evaluation criteria for the quality evaluation of the infrared simulation image.

[0006] The specific idea of ​​achieving the purpose of the present invention is that, in view of the problem that the prior art ignores the human visual characteristics, the present invention extracts the characteristics of image pixel brightness, image structure, image content, and image texture, and describes the characteristics of infrared simulation images from four different perspectives, so that the extracted characteristics can meet the visual characteristics of the human eye; in view of the problem that the evaluation index of each feature is manually weighted, the training set of infrared simulation image features is trained using a random forest algorithm, and subjective evaluation values ​​are used as labels, and different weights are given to each feature from the perspective of human subjective cognition, so as to obtain image quality evaluation results that are more consistent with human subjective perception;

[0007] To achieve the above purpose, the technical solution adopted by the invention includes the following steps:

[0008] Step 1, collecting infrared images, and simulating the collected infrared images to obtain infrared simulation images;

[0009] Step 2, using the peak signal-to-noise ratio formula to calculate the peak signal-to-noise ratio values ​​of the infrared image and the infrared simulation image, and using the peak signal-to-noise ratio value as the pixel brightness feature of the infrared simulation image;

[0010] Step 3, using a structural similarity formula to calculate the structural similarity between the infrared image and the infrared simulation image, and using the structural similarity value as a structural feature of the infrared simulation image;

[0011] Step 4: Extract shape features of infrared simulation images:

[0012] Step 4.1, using the Hu feature invariant moment formula, calculate the 7 Hu feature invariant moment values ​​of the infrared image shape, and compose the 7 Hu feature invariant moment values ​​into a shape feature vector;

[0013] Step 4.2, using the Hu feature invariant moment formula, calculate the 7 Hu feature invariant moment values ​​of the infrared simulation image shape, and compose the 7 Hu feature invariant moment values ​​into a shape feature vector;

[0014] Step 4.3, calculating the Euclidean distance between the infrared simulation image and the shape feature vector of the infrared image, and using the Euclidean distance as the shape feature of the infrared simulation image;

[0015] Step 5: Extract the texture features of the infrared simulation image:

[0016] Step 5.1, using Gabor transform formula, calculate the texture feature vector of the infrared image;

[0017] Step 5.2, using the Gabor transform formula, calculate the texture feature vector of the infrared simulation image;

[0018] Step 5.3, calculating the Euclidean distance between the infrared simulation image and the texture feature vector of the infrared image, and using the Euclidean distance as the texture feature of the infrared simulation image;

[0019] Step 6: Generate training set:

[0020] Step 6.1, randomly selecting 200 images from the infrared simulation image database containing infrared simulation targets to form a sample set, each sample including an infrared simulation image, an infrared image corresponding to the infrared simulation image, and a quality evaluation value;

[0021] Step 6.2, using the same processing method as steps 2 to 5, extracting features of an infrared simulation image and its corresponding infrared image in each sample from the sample set, and obtaining four features of image pixel brightness, image structure, image texture, and image shape of each infrared simulation image; the four features of each infrared simulation image in the sample set form a feature set of the infrared simulation image;

[0022] Step 6.3, the feature set of the sample set and the quality evaluation values ​​of the images corresponding to the features are combined into a training set;

[0023] Step 7, input the training set into the random forest model, use the random forest algorithm to train the training set, and obtain a trained random forest regressor model;

[0024] Step 8: Evaluate the quality of the infrared simulation image:

[0025] Step 8.1, using the same processing method as steps 2 to 5, to obtain four features of the infrared simulation image to be evaluated, namely, image pixel brightness, image structure, image texture, and image shape;

[0026] Step 8.2, input the four eigenvalues ​​into the random forest model at the same time, and use the random forest algorithm to obtain the quality evaluation value of the infrared simulation image to be evaluated.

[0027] Compared with the existing technology, it has the following advantages:

[0028] First, the present invention uses four features that can summarize the overall characteristics of infrared simulation images, namely, pixel brightness, image structure, image texture, and image shape, to form a training set. This overcomes the problem that the prior art ignores the feature information carried by image pixel brightness and image texture in the image, resulting in the evaluation result of the infrared simulation image not being consistent with human subjective perception. The quality evaluation result of the infrared simulation image evaluated by the present invention is more consistent with human subjective perception, and the infrared simulation image after quality evaluation by the present invention is more consistent with the fidelity of the infrared image.

[0029] Second, the present invention uses the random forest algorithm to train the training set of infrared simulation image features, and gives different weights to each feature from the perspective of human subjective cognition. This overcomes the prior art of artificially giving weights to each feature without considering the different proportions of different features in human vision, making the quality of the infrared simulation image evaluated by the present invention more consistent with the visual perception of the human eye. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0031] The following is combined with Figure 1 The steps for implementing the present invention are further described in detail with reference to the accompanying drawings and embodiments.

[0032] Step 1: Obtain infrared images and infrared simulation images.

[0033] Step 1.1, the embodiment of the present invention uses an infrared camera to photograph the car to obtain an infrared image with a size of 256*256;

[0034] Step 1.2, use the infrared simulation system to simulate the acquired infrared image to obtain an infrared simulation image with a size of 256*256.

[0035] Step 2: Calculate the peak signal-to-noise ratio of the infrared image and the infrared simulation image using the following formula as the pixel brightness feature of the infrared simulation image:

[0036]

[0037] Wherein, P represents the peak signal-to-noise ratio of the infrared image and the infrared simulation image, lg(·) represents the logarithmic operation with base 10, L represents the maximum value of the pixel when each sampling point in the infrared image and the infrared simulation image is 8 bits, and the value of the maximum value is 255, Σ represents the summation operation, n represents the total number of pixels in the infrared image or the infrared simulation image, and the value is 65536, c represents the sequence number of the pixel in the infrared image or the infrared simulation image, and I w represents the pixel value of the wth pixel in the infrared image, S v It represents the pixel value of the vth pixel in the infrared simulation image, and the values ​​of w and v are equal.

[0038] In the embodiment of the present invention, the evaluation results of the brightness similarity of infrared simulation images are divided into three types: "good", "medium" and "poor": when the peak signal-to-noise ratio value of the infrared image and the infrared simulation image is greater than 40dB, the brightness similarity quality of the infrared simulation image is judged as "good". When the peak signal-to-noise ratio value of the infrared image and the infrared simulation image is between 30-40dB, the brightness similarity quality of the infrared simulation image is judged as "medium". When the peak signal-to-noise ratio value of the infrared image and the infrared simulation image is between 20-30dB, the brightness similarity quality of the infrared simulation image is judged as "poor".

[0039] Step 3: Calculate the structural similarity value between the infrared image and the infrared simulation image using the following formula as the structural feature of the infrared simulation image:

[0040]

[0041] Among them, W(I,S) represents the structural similarity between the infrared image and the infrared simulation image, μ I Represents the average value of all pixel values ​​in the infrared image, μ S Represents the average value of all pixel values ​​in the infrared simulation image, C 1 Represents a constant that prevents the denominator from becoming smaller, C 1 =(0.01×L) 2 , σ IS represents the covariance of all corresponding pixel values ​​in the infrared image and the infrared simulation image, C 2 Represents a constant that prevents the denominator from becoming smaller, C 2 =(0.03×L) 2 , σ IRepresents the standard deviation of all pixel values ​​in the infrared image, σ S Represents the standard deviation of all pixel values ​​in the infrared simulation image.

[0042] In the embodiment of the present invention, when the structural similarity value between the infrared image and the infrared simulation image is closer to 1, it means that the infrared simulation image and the infrared image are more similar in structure, indicating that the structural similarity quality of the infrared simulation image is better.

[0043] Step 4: Calculate the Euclidean distance between the shape feature vectors of the infrared image and the infrared simulation image as the shape feature of the infrared simulation image.

[0044] Step 4.1, use the following formula to calculate the 7 Hu feature invariant moment values ​​of the infrared image shape, and combine the 7 Hu feature invariant moment values ​​into a shape feature vector:

[0045] M 1 =η 20 +η 02

[0046]

[0047] M 3 =(η 30 -3η 12 ) 2 +(3η 21 -η 03 ) 2

[0048] M 4 =(η 30 +η 12 ) 2 +(η 21 +η 03 ) 2

[0049] M 5 =(η 30 -3η 12 )(η 30 +η 12 )((η 30 +η 12 ) 2 -3(η 21 +η 03 ) 2 )+(3η 21 -η 03 )(η 21 +η 03 )(3(η 30 +η 12 ) 2 -(η 21 +η03 ) 2 )

[0050] M 6 =(η 20 -η 02 )((η 30 +η 12 ) 2 -(η 21 +η 03 ) 2 )+4η 11 (η 30 +η 12 )(η 21 +η 03 )

[0051] M 7 =(3η 21 -η 03 )(η 30 +η 12 )((η 30 +η 12 ) 2 -3(η 21 +η 03 ) 2 )-(η 30 -3η 12 )(η 21 +η 03 )(3(η 30 +η 12 ) 2 -(η 21 +η 03 ) 2 )

[0052] Among them, M 1 、M 2 、M 3 、M 4 、M 5 、M 6 、M 7 Respectively represent the 1st to 7th shape Hu feature invariant moment values ​​of infrared images, η 20 , η 02 Respectively represent the extension of the car in the horizontal and vertical directions in the infrared image, η 11 Indicates the inclination of the car direction in the infrared image, η 30 It represents the horizontal center of gravity deviation of the car outline in the infrared image, η 12 , η 21 Respectively represent the extension of the car in the vertical and horizontal directions in the infrared image, η 03 Indicates the vertical center of gravity deviation of the car outline in the infrared image.

[0053] Step 4.2, using the same method as step 4.1, obtain the shape feature vector of the infrared simulation image.

[0054] Step 4.3, calculate the Euclidean distance between the infrared simulation image and the infrared image shape feature vector using the following formula:

[0055]

[0056] Where D(I,S) represents the Euclidean distance between the infrared simulation image and the shape feature vector of the infrared image, Σ represents the summation operation, n represents the total number of shape feature values ​​in the shape feature vector of the infrared simulation image and the infrared image, and the value is 7, i represents the sequence number of the shape feature value in the shape feature vector of the infrared simulation image or the infrared image, and I m Represents the shape feature value in the shape feature vector of the infrared image, S w Represents the shape eigenvalue in the shape eigenvector of the infrared simulation image, and the values ​​of m and w are equal.

[0057] In the embodiment of the present invention, when the Euclidean distance between the shape Hu feature vectors of the infrared image and the infrared simulation image is smaller, it means that the infrared simulation image and the infrared image are more similar in shape, and the shape similarity quality of the infrared simulation image is better.

[0058] Step 5: Calculate the Euclidean distance between the texture feature vectors of the infrared image and the infrared simulation image as the texture feature of the infrared simulation image.

[0059] Step 5.1, use Gabor filtering to process the infrared image and the infrared simulation image to obtain the texture feature vectors of the infrared image and the infrared simulation image respectively.

[0060] Step 5.2, use the following formula to calculate the texture feature vector of the infrared image:

[0061]

[0062] Among them, g(x,y) represents the texture feature vector extracted from the infrared image after Gabor filtering, exp represents the exponential operation with the natural number e as the base, and x i represents the horizontal coordinate of the ith pixel in the infrared image, γ represents the aspect ratio of the kernel function of the Gabor filter used in the infrared image, and its value is 0.5, y i represents the ordinate of the i-th pixel in the infrared image, σ represents the standard deviation of the Gaussian function of the Gabor filter used in the infrared image, and the value is i represents the sign of the imaginary part, λ represents the wavelength of the Gabor filter used in the infrared image, and the value is 5, and ψ represents the phase offset of the cosine function in the kernel function of the Gabor filter used in the infrared image.

[0063] Step 5.3, use the same method as step 5.2 to obtain the texture feature vector of the infrared simulation image.

[0064] Step 5.4, calculate the Euclidean distance between the texture feature vectors of the infrared simulation image and the infrared image using the following formula:

[0065]

[0066] Where D(I,S) represents the Euclidean distance between the texture feature vectors of the infrared simulation image and the infrared image. represents a square root operation, Σ represents a summation operation, n represents the total number of eigenvalues ​​of the texture feature vector of the infrared image or the infrared simulation image, and in the embodiment of the present invention, n=40, i represents the sequence number of the eigenvalue in the texture feature vector of the infrared image or the infrared simulation image, and I b represents the bth eigenvalue in the texture feature vector of the infrared image, S a It represents the ath eigenvalue in the texture feature vector of the infrared simulation image, and the values ​​of b and a are equal.

[0067] In the embodiment of the present invention, when the Euclidean distance between the texture feature vectors of the infrared image and the infrared simulation image is smaller, it means that the infrared simulation image and the infrared image are more similar in texture, which means that the texture similarity quality of the infrared simulation image is better.

[0068] Step 6: Generate a training set.

[0069] Step 6.1, in an embodiment of the present invention, 200 infrared simulation image samples are randomly selected from an infrared simulation image database of an infrared simulation target vehicle to form a sample set, each sample comprising an infrared simulation image, an infrared image corresponding to the infrared simulation image, and a quality evaluation value;

[0070] Step 6.2, using the same processing method as steps 2 to 5, extract the features of an infrared simulation image and its corresponding infrared image in each sample from the sample set, and obtain four features of image pixel brightness, image structure, image texture, and image shape of each infrared simulation image. The four features of each infrared simulation image in the sample set form a feature set of the infrared simulation image.

[0071] Step 6.3, the feature set of the sample set and the quality evaluation values ​​of the images corresponding to the features form a training set.

[0072] Step 7: Train the random forest model.

[0073] The training set is input into the random forest model, and the training set is trained using the random forest algorithm to obtain a trained random forest regressor model.

[0074] Step 8: Evaluate the quality of the infrared simulation image.

[0075] Step 8.1, using the same processing method as steps 2 to 5, to obtain four features of the infrared simulation image to be evaluated, namely, image pixel brightness, image structure, image texture, and image shape;

[0076] Step 8.2, input the four eigenvalues ​​into the random forest model at the same time, and use the random forest algorithm to obtain the quality evaluation value of the infrared simulation image to be evaluated.

Claims

1. A method for evaluating infrared simulation image quality based on random forest, characterized in that: The four features of pixel brightness, image structure, image texture and image shape are used to form a training set, and the quality of infrared simulation images is evaluated using the trained random forest model. The specific steps of this method include the following: Step 1, collecting infrared images, and simulating the collected infrared images to obtain infrared simulation images; Step 2, using the peak signal-to-noise ratio formula, calculate the peak signal-to-noise ratio values ​​of the infrared image and the infrared simulation image, and use the peak signal-to-noise ratio value as the pixel brightness feature of the infrared simulation image; Step 3, using a structural similarity formula to calculate the structural similarity between the infrared image and the infrared simulation image, and using the structural similarity value as a structural feature of the infrared simulation image; Step 4: Extract shape features of infrared simulation images: Step 4.1, using the Hu feature invariant moment formula, calculate the 7 Hu feature invariant moment values ​​of the infrared image shape, and combine the 7 Hu feature invariant moment values ​​into a shape feature vector; Step 4.2, using the Hu feature invariant moment formula, calculate the 7 Hu feature invariant moment values ​​of the infrared simulation image shape, and compose the 7 Hu feature invariant moment values ​​into a shape feature vector; Step 4.3, calculating the Euclidean distance between the infrared simulation image and the shape feature vector of the infrared image, and using the Euclidean distance as the shape feature of the infrared simulation image; Step 5: Extract the texture features of the infrared simulation image: Step 5.1, using Gabor transform formula, calculate the texture feature vector of the infrared image; Step 5.2, using the Gabor transform formula, calculate the texture feature vector of the infrared simulation image; Step 5.3, calculating the Euclidean distance between the infrared simulation image and the texture feature vector of the infrared image, and using the Euclidean distance as the texture feature of the infrared simulation image; Step 6: Generate training set: Step 6.1, randomly selecting 200 images from the infrared simulation image database containing infrared simulation targets to form a sample set, each sample including an infrared simulation image, an infrared image corresponding to the infrared simulation image, and a quality evaluation value; Step 6.2, using the same processing method as steps 2 to 5, extracting features of an infrared simulation image and its corresponding infrared image in each sample from the sample set, and obtaining four features of image pixel brightness, image structure, image texture, and image shape of each infrared simulation image; the four features of each infrared simulation image in the sample set form a feature set of the infrared simulation image; Step 6.3, the feature set of the sample set and the quality evaluation values ​​of the images corresponding to the features are combined into a training set; Step 7, input the training set into the random forest model, use the random forest algorithm to train the training set, and obtain a trained random forest regressor model; Step 8: Evaluate the quality of the infrared simulation image: Step 8.1, using the same processing method as steps 2 to 5, to obtain four features of the infrared simulation image to be evaluated, namely, image pixel brightness, image structure, image texture, and image shape; Step 8.2, input the four eigenvalues ​​into the random forest model at the same time, and use the random forest algorithm to obtain the quality evaluation value of the infrared simulation image to be evaluated.

2. The infrared simulation image quality evaluation method based on random forest according to claim 1 is characterized in that: The peak signal-to-noise ratio formula described in step 2 is as follows: Where P represents the peak signal-to-noise ratio of the infrared image and the infrared simulation image, lg(·) represents the logarithmic operation with base 10, L represents the maximum value of the pixel when each sampling point in the infrared image and the infrared simulation image is 8 bits, and the value of the maximum value is 255, Σ represents the summation operation, n represents the total number of pixels in the infrared image or the infrared simulation image, c represents the sequence number of the pixel in the infrared image or the infrared simulation image, and I w represents the pixel value of the wth pixel in the infrared image, S v It represents the pixel value of the vth pixel in the infrared simulation image, and the values ​​of w and v are equal.

3. The infrared simulation image quality evaluation method based on random forest according to claim 1 is characterized in that: The structural similarity formula described in step 3 is as follows: Among them, W(I,S) represents the structural similarity between the infrared image and the infrared simulation image, μ I Represents the average value of all pixel values ​​in the infrared image, μ S Represents the average value of all pixel values ​​in the infrared simulation image, C1 represents a constant to prevent the denominator from becoming smaller, C1=(0.01×L) 2 , σ IS Represents the covariance of all corresponding pixel values ​​in the infrared image and the infrared simulation image. C2 represents a constant to prevent the denominator from becoming smaller. C2=(0.03×L) 2 , σ I Represents the standard deviation of all pixel values ​​in the infrared image, σ S Represents the standard deviation of all pixel values ​​in the infrared simulation image.

4. The infrared simulation image quality evaluation method based on random forest according to claim 1 is characterized in that: The Hu characteristic invariant moment formula described in step 4 is as follows: M1=η 20 +n 02 M3=(η 30 -3rd 12 ) 2 +(3rd 21 -or 03 ) 2 M4=(η 30 +n 12 ) 2 +(the 21 +n 03 ) 2 M5=(η 30 -3rd 12 )(or 30 +n 12 )((or 30 +n 12 ) 2 -3(h 21 +n 03 ) 2 )+(3rd 21 -or 03 )(or 21 +n 03 )(3(h 30 +n 12 ) 2 -(or 21 +n 03 ) 2 ) M6=(η 20 -or 02 )((or 30 +n 12 ) 2 -(or 21 +n 03 ) 2 )+4th 11 (or 30 +n 12 )(or 21 +n 03 ) M7=(3rd 21 -or 03 )(or 30 +n 12 )((or 30 +n 12 ) 2 -3(h 21 +n 03 ) 2 )-(or 30 -3rd 12 )(or 21 +n 03 )(3(h 30 +n 12 ) 2 -(or 21 +n 03 ) 2 ) Among them, M1, M2, M3, M4, M5, M6, and M7 represent the 1st to 7th shape Hu feature invariant moments of the infrared image, respectively, and η 20 , η 02 Respectively represent the stretch of the image content in the infrared image in the horizontal and vertical directions, η 11 Indicates the tilt of the direction of the image content in the infrared image, η 30 Indicates the horizontal center of gravity deviation of the contour of the image content in the infrared image, η 12 , η 21 Respectively represent the stretch of the image content in the infrared image in the vertical and horizontal directions, η 03 Indicates the vertical center of gravity deviation of the outline of the image content in the infrared image.

5. The infrared simulation image quality evaluation method based on random forest according to claim 1 is characterized in that: The Gabor transform formula described in step 5 is as follows: Among them, g(x,y) represents the texture feature vector extracted from the infrared image after Gabor filtering, exp represents the exponential operation with the natural number e as the base, and x i represents the horizontal coordinate of the i-th pixel in the infrared image, γ represents the aspect ratio of the kernel function of the Gabor filter used in the infrared image, and y i represents the ordinate of the i-th pixel in the infrared image, σ represents the standard deviation of the Gaussian function of the Gabor filter used in the infrared image, i represents the sign of the imaginary part, λ represents the wavelength of the Gabor filter used in the infrared image, and ψ represents the phase offset of the cosine function in the kernel function of the Gabor filter used in the infrared image.

Citation Information

Patent Citations

  • Image quality judgment method and device and storage medium

    CN111798421A

  • No-reference structure definition evaluation method based on human vision system

    CN113570596A

  • An image quality evaluation method based on integrated learning and a random forest

    CN109919901A

  • No-reference image quality evaluation method based on deep forest classification

    CN110400293A

  • Quality evaluation of sequences of images

    EP2396768A1