Two-dimensional discrete wave-based dual-frequency disproportionation deconstruction method for military polarization image

Abstract

The invention relates to the technical field of military target information image fusion processing, and provides a dual-frequency disproportionation deconstruction method for a military polarization image based on a two-dimensional discrete wave, and the method comprises the following steps: S1, obtaining an image; step S2, primary treatment; s3, fusing the degree of polarization; step S4, decomposition pretreatment; and S5, wavelet decomposition and reconstruction: carrying out Dolp-Aop decomposition on the light intensity image and the polarization characteristic image to obtain a low-frequency component and a high-frequency component, and obtaining a final fusion image by adopting a specific processing mode. Different processing methods are adopted for wavelet decomposition coefficients of different frequency bands obtained after wavelet decomposition, the wavelet coefficients obtained through decomposition are processed through a weighted average method, the frequency characteristics are reserved as far as possible, the true structural characteristics of an image are reflected, and it is guaranteed that information obtained after information fusion is true and accurate.

Description

Technical Field

[0001] The present invention relates to the technical field of military target information image fusion processing, and in particular to a dual-frequency disambiguation deconstruction method for military polarization images based on two-dimensional discrete waves. Background Art

[0002] With the advancement of technology, polarization effects in the transmission of light waves such as infrared reflection, thermal radiation and scattering have been deeply analyzed and modeled, and have been experimentally verified. In recent years: polarization imaging technology has developed rapidly and has been widely used in many fields. For example, in environmental monitoring, polarization imaging technology can be used to detect atmospheric pollutants; in the field of remote sensing, polarization imaging technology can improve the accuracy of ground object recognition; in security monitoring, polarization imaging technology can be used for camouflage recognition and defogging imaging, etc.

[0003] Image fusion is a branch of data fusion that mainly studies images. It means to integrate the images or image sequence information of a specific scene acquired by two or more sensors at the same time or at different times to generate new information processing about the scene interpretation. Images are captured using polarization cameras or other polarization imaging devices, and the acquired polarization images are processed to calculate the type, orientation, speed and other information of the expected target. Making full use of this accurate information is of great value for command and decision support and precise strike targets.

[0004] Therefore, taking advantage of the information richness, material identification, multi-field adaptability, non-contact and non-destructive testing of polarized images, we propose a dual-frequency disambiguation deconstruction method for military polarized images based on two-dimensional discrete waves, considering that traditional imaging methods are easily affected by complex battlefield environments and there are few image fusion processing methods that can obtain more information about military targets by processing and fusing polarized images. By processing and fusing the intensity image and polarization image of military targets, a fused image with higher information content is obtained, which helps people obtain more information about military targets in military reconnaissance.

[0005] In order to preserve the information characteristics of the target to the maximum extent, that is, to preserve the polarization degree (D olp ) image and polarization angle (A op ) image details information, we use the Toet image fusion method to fuse it.

[0006] like Figure 1As shown in the figure, the traditional Toet image fusion method (Thresholded Overlapping Eigenvector Transform) is a technology based on the Ratio of Low-Pass Pyramid (RP), which can fuse the information of multiple input images into a high-quality fused image. We make adaptive modifications to the traditional Toet image fusion method to make it better meet our needs.

[0007] The principle of the original Toet image fusion method is: first convert the input image into a grayscale image so that each pixel can be processed. Then preprocess the grayscale image to reduce the influence of noise. Next, calculate the gradient image of the image, that is, calculate the gradient value of each pixel to extract the edge information of the image. Then, perform threshold segmentation on the gradient image to obtain a binary image. Finally, fuse the binary image with the original image to obtain the final fused image. Summary of the invention

[0008] The present invention proposes a dual-frequency disambiguation deconstruction method for military polarization images based on two-dimensional discrete waves, which solves the problem that traditional imaging methods in related technologies are easily affected by complex battlefield environments.

[0009] The technical solution of the present invention is as follows: A dual-frequency disambiguation deconstruction method for military polarization images based on two-dimensional discrete waves includes the following steps:

[0010] Step S1, image acquisition: using an ordinary camera to acquire a light intensity image, by adding a polarizing plate to the ordinary camera to acquire a polarized image, and rotating the polarizing plate in front of the polarization camera to acquire polarized light intensity images in directions forming four angles of 0°, 45°, 90° and 135° with the horizontal direction;

[0011] Step S2: Preliminary processing: using the formula of Stokes vector

[0012] The three component images I, Q and U are obtained respectively;

[0013] According to the linear polarization

[0014] And polarization angle calculation formula

[0015] The three component images I, Q, and U are processed and fused to obtain the linear polarization degree (D olp ) and polarization angle (A op )image;

[0016] Step S3: Polarization degree fusion: In order to preserve the linear polarization degree (D olp) and polarization angle (A op ) The information features on the image are firstly calculated using the formula Calculate D olp (x, y) and A op (x, y), and then according to the modified Toet image fusion algorithm Right D olp (x, y) and A op (x, y) is processed to obtain the polarization characteristic image D olp -A OP ;

[0017] Step S4, decomposition preprocessing: In order to perform wavelet decomposition on the fused image, the intensity image and polarization feature image D are first obtained. olp -A OP Perform registration processing to make the two match in terms of avatar pixels;

[0018] Step S5, wavelet decomposition and reconstruction: First, the intensity image and polarization feature image D olp -A OP After dual-frequency disparity deconstruction, the intensity image and polarization feature image D olp -A OP The low-frequency components and high-frequency components obtained by decomposition are processed in a specific way to obtain the final fused image.

[0019] Preferably, in step S2, the formula Ia (a=0°, 45°, 90°, 135°) of the Stokes vector represents the polarization image a corresponding to different polarization directions;

[0020] S0 or I represents the total light intensity, that is, the sum of the light intensities in all polarization directions, which reflects the overall brightness or energy of the light beam;

[0021] S1 or Q represents the intensity difference between the linear polarized light component in the horizontal direction and the linear polarized light component in the vertical direction, which reflects the difference in polarization degree of the light beam in the horizontal and vertical directions;

[0022] S2 or U represents the intensity difference between the linear polarized light component at 45° and the linear polarized light component at -45° or 135°, which reveals the polarization characteristics of the light beam in the diagonal direction;

[0023] S3 or V represents the intensity difference between the right-handed circularly polarized light component and the left-handed circularly polarized light component, which reflects the circular polarization component in the light beam, that is, whether the photons in the light beam rotate right or left.

[0024] Preferably, in step S2, the linear polarization degree (D olp ) represents the proportion of linear polarized light intensity in the total light intensity, and the polarization angle (A op) represents the angle between the vibration direction of linear polarized light and the horizontal direction.

[0025] Preferably, in step S3, a, b and c are all weight factors;

[0026] In the simulation experiment, the best simulation effect is achieved when a, b and c are 0.6, 0.2 and 0.2 respectively;

[0027] f(x, y) is D olp With A OP The fused polarization feature image D olp -A OP .

[0028] Preferably, in step S5, the intensity image and the polarization characteristic image D olp -A OP The dual-frequency disparity deconstruction method is as follows: the intensity image and the polarization feature image D olp -A OP Perform discrete wavelet decomposition in two dimensions.

[0029] Preferably, in step S5, the intensity image and the polarization characteristic image D olp -A OP The low-frequency components and high-frequency components obtained by decomposition are processed in a specific way: for the low-frequency components, in order to retain the frequency characteristics and reflect the true characteristics of the image, the low-frequency wavelet coefficients obtained by decomposition are processed by weighted average method to make the information obtained after the fusion of information true and accurate; for the high-frequency components, in order to retain the intensity image and polarization characteristic image D olp -A OP The image structure attributes are processed by an improved fusion method based on the Sobel operator to reduce image noise and make the fused image smoother. Finally, the intensity image and polarization feature image A are combined. olp -A OP The low-frequency component and high-frequency component after fusion processing are reconstructed by two-dimensional discrete wavelet to obtain the final fused image.

[0030] Preferably, the Sobel operator includes two groups of 3x3 matrices, one for horizontal direction and one for vertical direction;

[0031] The fusion method of the Sobel operator is as follows: a) performing planar convolution of the Sobel operator and the image, and the calculation equation is:

[0032]

[0033] b) Obtain the approximate values ​​of the horizontal and vertical brightness differences, with A representing the original image and G x and G yRespectively represent the images after horizontal and vertical edge detection, and the formulas are as follows:

[0034]

[0035] c) The approximate values ​​of the horizontal and vertical gradients of each pixel in the image can be expressed as Combined to calculate the magnitude of the gradient;

[0036] d) Then the following formula can be used Calculate the gradient direction, and realize edge detection by calculating the gradient of the image in the x-direction and y-direction and combining the magnitude and direction information of the gradient.

[0037] Preferably, in step S5, the specific method of wavelet decomposition and reconstruction is: in a discrete case, a pyramid algorithm of discrete wavelet transform is obtained based on the Mallat algorithm using the concept of multi-resolution analysis, and the original input signal is c0, then the algorithm can be described as:

[0038] Preferably, in the pyramid algorithm of wavelet decomposition and reconstruction, h and g are the low-pass and high-pass wavelet filters of the decomposition, For low frequency analysis signal, It is called high-frequency detail signal, J is the transformation level, ≤log2N, N is the number of samples of the original input signal;

[0039] The algorithm can be extended to the case of two-dimensional image signals with separable filters. The specific method is as follows: first, each row of the image is transformed, and then each column is transformed to obtain four sub-band images representing different frequency and direction characteristics of the original image, which are denoted as LL, HL, LH, and HH respectively;

[0040] Among them, LL is the low-frequency subband where the energy is concentrated, reflecting the change of grayscale;

[0041] HL is the horizontal high-frequency and vertical low-frequency subbands with vertical edge information;

[0042] LH is the horizontal low-frequency and vertical high-frequency subband with horizontal edge information;

[0043] HH is the horizontal low-frequency vertical high-frequency subband with diagonal edge information;

[0044] The LL subband can be further decomposed into four subbands to complete a multi-level two-dimensional wavelet transform.

[0045] The working principle and beneficial effects of the present invention are:

[0046] 1. The present invention adopts different processing methods for the wavelet decomposition coefficients of different frequency bands obtained after wavelet decomposition, and the decomposed wavelet coefficients are processed by weighted average method, so as to retain the frequency characteristics as much as possible, reflect the true structural characteristics of the image, and ensure the authenticity and accuracy of the information obtained after the fusion of information.

[0047] 2. By processing the high-frequency components using a fusion method based on the Sobel operator, image noise can be reduced, making the fused image smoother and ensuring the clarity of the fused image.

[0048] 3. Direct observation in the present invention shows that the fusion method of weighted average, small absolute value and Prewitt operator makes the final fused image have lower brightness, especially the fusion method of small absolute value and Prewitt operator, which greatly affects the manual reading of the target information. The fused image obtained by the method in this paper has great advantages in viewing and readability compared with the fused images obtained by other methods.

[0049] 4. The fused image obtained by the method used in the present invention is optimal in the evaluation of information entropy and standard deviation, and is superior to the weighted average method and the minimum absolute value method in terms of average gradient. The fused image obtained by the algorithm in this paper has high contrast, contains rich information, and can provide more detailed information of the target. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0051] Figure 1 This is the traditional flow chart of Toet image fusion method;

[0052] Figure 2 It is a flow chart of the improved Toet image fusion method in the present invention;

[0053] Figure 3 It is a two-dimensional discrete wavelet decomposition flow chart in the present invention;

[0054] Figure 4 It is a diagram of the two-dimensional discrete wavelet decomposition image transformation process in the present invention;

[0055] Figure 5 It is a two-dimensional discrete wavelet decomposition flow chart adopted in the present invention;

[0056] Figure 6 is a comprehensive flow chart of the method of the present invention;

[0057] Figure 7 The polarization image diagrams of the four polarization degrees of 0°, 45°, 90° and 135° required for collecting in the present invention;

[0058] Figure 8is the fused image of component I in the present invention;

[0059] Fig. 9 It is a fused image diagram of the Q component in the present invention;

[0060] Fig.10 is the fused image of the U component in the present invention;

[0061] Fig.11 is the linear polarization degree D in the present invention olp Image graph of

[0062] Fig.12 is the polarization angle A in the present invention op Image graph of

[0063] Fig.13 is the final fused image obtained in the present invention;

[0064] Fig.14 It is the information index diagram of the present invention;

[0065] Fig.15 is the final fused image obtained in the first embodiment;

[0066] Fig.16 This is the final fused image obtained in the second embodiment;

[0067] Fig.17 This is the final fused image obtained in the third embodiment;

[0068] Fig.18 This is the final fused image obtained in the fourth embodiment. DETAILED DESCRIPTION

[0069] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0070] Embodiment 1

[0071] like Figure 1-Figure 14As shown, this embodiment proposes a dual-frequency dissimilarity deconstruction method for military polarization images based on two-dimensional discrete waves. The codes of this experiment are all implemented by running Python3.8.10, and the shooting parameters are: f / 1.61 / 50s ISO500 equivalent focal length 24mm3072*3072p; the steps are as follows: Step S1, image acquisition: use an ordinary camera to acquire a light intensity image, and obtain a polarization image by adding a polarizer to the ordinary camera, and rotate the polarizer in front of the polarization camera to collect polarized light intensity images in directions that form four angles of 0°, 45°, 90° and 135° with the horizontal direction;

[0072] Step S2: Preliminary processing: using the formula of Stokes vector

[0073] The three component images I, Q and U are obtained respectively;

[0074] According to the linear polarization

[0075] And polarization angle calculation formula

[0076] The three component images I, Q, and U are processed and fused to obtain the linear polarization degree (D olp ) and polarization angle (A op ) image, the formula of Stokes vector Ia (a = 0°, 45°, 90°, 135°) represents the polarization image a corresponding to different polarization directions; S0 or I represents the total light intensity of the light, that is, the sum of the light intensities in all polarization directions, which reflects the overall brightness or energy of the light beam; S1 or Q represents the intensity difference between the linear polarized light component in the horizontal direction and the linear polarized light component in the vertical direction, which reflects the difference in polarization degree of the light beam in the horizontal and vertical directions; S2 or U represents the intensity difference between the linear polarized light component in the 45° direction and the linear polarized light component in the -45° direction or 135° direction, which reveals the polarization characteristics of the light beam in the diagonal direction; S3 or V represents the intensity difference between the right-handed circularly polarized light component and the left-handed circularly polarized light component, which reflects the circularly polarized component in the light beam, that is, whether the photons in the light beam rotate to the right or to the left; the degree of linear polarization (D olp ) represents the proportion of linear polarized light intensity in the total light intensity, and the polarization angle (A op ) represents the angle between the vibration direction of linear polarized light and the horizontal direction;

[0077] Step S3: Polarization degree fusion: In order to preserve the linear polarization degree (D olp ) and polarization angle (A op ) The information features on the image are firstly calculated using the formula Calculate D olp (x, y) and A op(x, y), and then according to the modified Toet image fusion algorithm Right D olp (x, y) and A op (x, y) is processed to obtain the polarization characteristic image D olp -A OP ; a, b and c are weight factors;

[0078] In the simulation experiment, the best simulation effect is achieved when a, b and c are 0.6, 0.2 and 0.2 respectively;

[0079] f(x, y) is D olp With A OP The fused polarization feature image D olp -A OP ;

[0080] Step S4, decomposition preprocessing: In order to perform wavelet decomposition on the fused image, the intensity image and polarization feature image D are first obtained. olp -A OP Perform registration processing to make the two match in terms of avatar pixels;

[0081] Step S5, wavelet decomposition and reconstruction: First, the intensity image and polarization feature image D olp -A OP Perform dual-frequency disparity deconstruction processing to the intensity image and polarization feature image D olp -A OP The dual-frequency disparity deconstruction method is as follows: the intensity image and the polarization feature image D olp -A OP Perform discrete wavelet decomposition on a two-dimensional level, and then perform the intensity image and polarization feature image D olp -A OP The low-frequency components and high-frequency components obtained by decomposition are processed in a specific way to obtain the final fused image. olp -A OP The low-frequency components and high-frequency components obtained by decomposition are processed in a specific way: for the low-frequency components, in order to retain the frequency characteristics and reflect the true characteristics of the image, the low-frequency wavelet coefficients obtained by decomposition are processed by weighted average method to make the information obtained after the fusion of information true and accurate; for the high-frequency components, in order to retain the intensity image and polarization characteristic image D olp -A OP The image structure attributes are processed by an improved fusion method based on the Sobel operator to reduce image noise and make the fused image smoother. Finally, the intensity image and polarization feature image D olp -A OPThe low-frequency component and high-frequency component after fusion processing are reconstructed by two-dimensional discrete wavelet to obtain the final fused image. The Sobel operator contains two groups of 3x3 matrices, one for the horizontal direction and the other for the vertical direction.

[0082] The fusion method of the Sobel operator is as follows: a) performing planar convolution of the Sobel operator and the image, and the calculation equation is:

[0083]

[0084] b) Obtain the approximate values ​​of the horizontal and vertical brightness differences, with A representing the original image and G x and G y Respectively represent the images after horizontal and vertical edge detection, and the formulas are as follows:

[0085]

[0086] c) The approximate values ​​of the horizontal and vertical gradients of each pixel in the image can be expressed as Combined to calculate the magnitude of the gradient;

[0087] d) Then the following formula can be used Calculate the gradient direction, by calculating the gradient of the image in the x and y directions, and combining the amplitude and direction information of the gradient, to achieve edge detection. The specific method of wavelet decomposition and reconstruction is: In the discrete case, based on the Mallat algorithm, the concept of multi-resolution analysis is used to derive the pyramid algorithm of discrete wavelet transform. Let the original input signal be c0, then the algorithm can be described as: In the pyramid algorithm of wavelet decomposition and reconstruction, h and g are the low-pass and high-pass wavelet filters of the decomposition. For low frequency analysis signal, It is called the high-frequency detail signal, J is the transformation order, ≤log2N, N is the number of samples of the original input signal; the algorithm can be extended to the case of a two-dimensional image signal with a separable filter, and the specific method is as follows: first transform each row of the image, and then transform each column to obtain four sub-band images representing different frequency and directional characteristics of the original image, which are represented as LL, HL, LH, and HH respectively; among them, LL is a low-frequency sub-band with concentrated energy, reflecting the change of grayscale; HL is a horizontal high-frequency and vertical low-frequency sub-band with vertical edge information; LH is a horizontal low-frequency and vertical high-frequency sub-band with horizontal edge information; HH is a horizontal low-frequency and vertical high-frequency sub-band with diagonal edge information; the LL sub-band can be further decomposed into 4 sub-bands to complete a multi-level two-dimensional wavelet transform.

[0088] Embodiment 2

[0089] like Figure 1-Figure 14As shown, this embodiment proposes a dual-frequency dissimilarity deconstruction method for military polarization images based on two-dimensional discrete waves. The codes of this experiment are all implemented by running Python3.8.10, and the shooting parameters are: f / 1.61 / 50s ISO500 equivalent focal length 24mm3072*3072p; the steps are as follows: Step S1, image acquisition: use an ordinary camera to acquire a light intensity image, and obtain a polarization image by adding a polarizer to the ordinary camera, and rotate the polarizer in front of the polarization camera to collect polarized light intensity images in directions that form four angles of 0°, 45°, 90° and 135° with the horizontal direction;

[0090] Step S2: Preliminary processing: using the formula of Stokes vector

[0091] The three component images I, Q and U are obtained respectively;

[0092] According to the linear polarization

[0093] And polarization angle calculation formula

[0094] The three component images I, Q, and U are processed and fused to obtain the linear polarization degree (D olp ) and polarization angle (A op ) image, the formula of Stokes vector Ia (a = 0°, 45°, 90°, 135°) represents the polarization image a corresponding to different polarization directions; S0 or I represents the total light intensity of the light, that is, the sum of the light intensities in all polarization directions, which reflects the overall brightness or energy of the light beam; S1 or Q represents the intensity difference between the linear polarized light component in the horizontal direction and the linear polarized light component in the vertical direction, which reflects the difference in polarization degree of the light beam in the horizontal and vertical directions; S2 or U represents the intensity difference between the linear polarized light component in the 45° direction and the linear polarized light component in the -45° direction or 135° direction, which reveals the polarization characteristics of the light beam in the diagonal direction; S3 or V represents the intensity difference between the right-handed circularly polarized light component and the left-handed circularly polarized light component, which reflects the circularly polarized component in the light beam, that is, whether the photons in the light beam rotate to the right or to the left; the degree of linear polarization (D olp ) represents the proportion of linear polarized light intensity in the total light intensity, and the polarization angle (A op ) represents the angle between the vibration direction of linear polarized light and the horizontal direction;

[0095] Step S3: using weighted average method to obtain the final fused image.

[0096] Embodiment 3

[0097] like Figure 1-Figure 14As shown, this embodiment proposes a dual-frequency dissimilarity deconstruction method for military polarization images based on two-dimensional discrete waves. The codes of this experiment are all implemented by running Python3.8.10, and the shooting parameters are: f / 1.61 / 50s ISO500 equivalent focal length 24mm3072*3072p; the steps are as follows: Step S1, image acquisition: use an ordinary camera to acquire a light intensity image, and obtain a polarization image by adding a polarizer to the ordinary camera, and rotate the polarizer in front of the polarization camera to collect polarized light intensity images in directions that form four angles of 0°, 45°, 90° and 135° with the horizontal direction;

[0098] Step S2: Preliminary processing: using the formula of Stokes vector

[0099] The three component images I, Q and U are obtained respectively;

[0100] According to the linear polarization

[0101] And polarization angle calculation formula

[0102] The three component images I, Q, and U are processed and fused to obtain the linear polarization degree (D olp ) and polarization angle (A op ) image, the formula of Stokes vector Ia (a = 0°, 45°, 90°, 135°) represents the polarization image a corresponding to different polarization directions; S0 or I represents the total light intensity of the light, that is, the sum of the light intensities in all polarization directions, which reflects the overall brightness or energy of the light beam; S1 or Q represents the intensity difference between the linear polarized light component in the horizontal direction and the linear polarized light component in the vertical direction, which reflects the difference in polarization degree of the light beam in the horizontal and vertical directions; S2 or U represents the intensity difference between the linear polarized light component in the 45° direction and the linear polarized light component in the -45° direction or 135° direction, which reveals the polarization characteristics of the light beam in the diagonal direction; S3 or V represents the intensity difference between the right-handed circularly polarized light component and the left-handed circularly polarized light component, which reflects the circularly polarized component in the light beam, that is, whether the photons in the light beam rotate to the right or to the left; the degree of linear polarization (D olp ) represents the proportion of linear polarized light intensity in the total light intensity, and the polarization angle (A op ) represents the angle between the vibration direction of linear polarized light and the horizontal direction;

[0103] Step S3: Obtain the final fused image using the minimum absolute value method.

[0104] Embodiment 4

[0105] like Figure 1-Figure 14As shown, this embodiment proposes a dual-frequency dissimilarity deconstruction method for military polarization images based on two-dimensional discrete waves. The codes of this experiment are all implemented by running Python3.8.10, and the shooting parameters are: f / 1.61 / 50s ISO500 equivalent focal length 24mm3072*3072p; the steps are as follows: Step S1, image acquisition: use an ordinary camera to acquire a light intensity image, and obtain a polarization image by adding a polarizer to the ordinary camera, and rotate the polarizer in front of the polarization camera to collect polarized light intensity images in directions that form four angles of 0°, 45°, 90° and 135° with the horizontal direction;

[0106] Step S2: Preliminary processing: using the formula of Stokes vector

[0107] The three component images I, Q and U are obtained respectively;

[0108] According to the linear polarization

[0109] And polarization angle calculation formula

[0110] The three component images I, Q, and U are processed and fused to obtain the linear polarization degree (D olp ) and polarization angle (A op ) image, the formula of Stokes vector Ia (a = 0°, 45°, 90°, 135°) represents the polarization image a corresponding to different polarization directions; S0 or I represents the total light intensity of the light, that is, the sum of the light intensities in all polarization directions, which reflects the overall brightness or energy of the light beam; S1 or Q represents the intensity difference between the linear polarized light component in the horizontal direction and the linear polarized light component in the vertical direction, which reflects the difference in polarization degree of the light beam in the horizontal and vertical directions; S2 or U represents the intensity difference between the linear polarized light component in the 45° direction and the linear polarized light component in the -45° direction or 135° direction, which reveals the polarization characteristics of the light beam in the diagonal direction; S3 or V represents the intensity difference between the right-handed circularly polarized light component and the left-handed circularly polarized light component, which reflects the circularly polarized component in the light beam, that is, whether the photons in the light beam rotate to the right or to the left; the degree of linear polarization (D olp ) represents the proportion of linear polarized light intensity in the total light intensity, and the polarization angle (A op ) represents the angle between the vibration direction of linear polarized light and the horizontal direction;

[0111] Step S3: using the prewitt operator method to obtain the final fused image.

[0112] The fused images obtained in the above four embodiments are compared subjectively and objectively:

[0113] The present invention aims at the fact that traditional imaging methods are easily affected by the complex battlefield environment, and there are currently few image fusion processing methods that can obtain more information about military targets by processing and fusing polarization images. Therefore, the final fused image will not only be transmitted to a computer for analysis, but also be directly read by humans, so the evaluation system is divided into subjective comparison and objective comparison.

[0114] Subjective comparison: The subjective evaluation standard of fused image quality is mainly based on people's visual comfort and perception. It is a qualitative evaluation standard. The evaluation method is relatively simple in form and is mainly based on the evaluator. The subjective feelings of different evaluators may be slightly different, and it is not possible to be completely objective and fair. Therefore, only a brief explanation is given here. Direct observation shows that the fusion method of weighted average, small absolute value, and Prewitt operator makes the final fused image lower in brightness. Figure 15-18 , especially the fusion method with small absolute value and Prewitt operator, which greatly affects the manual reading of target information. The fusion image obtained in Example 1 has greater advantages in viewing and readability than the fusion images obtained by other methods.

[0115] Objective comparison: Since there is no unified evaluation index for fused images, we evaluate the fused images from the perspective of information entropy, average gradient, spatial frequency, and standard deviation, and perform the following calculations on the data of Embodiments 1, 2, 3, and 4:

[0116] Information entropy is a measure of the richness of image information. The greater the information entropy, the more information the image contains. The calculation formula of information entropy is usually expressed as:

[0117]

[0118] Among them, p(g) is the ratio of the number of pixels with grayscale equal to g to the total number of pixels in the image, and satisfies This formula calculates the probability of each gray level appearing in the image and uses these probabilities to evaluate the information content of the image.

[0119] The average gradient is mainly used to measure the clarity of an image. It can sensitively reflect the image's ability to express the contrast of tiny details and at the same time reflect the characteristics of texture transformation in the image. The specific formula is:

[0120]

[0121] Among them, H(i,j) represents the grayscale value of the image at point (i,j), and M and N are the total number of rows and columns of the image respectively.

[0122] Spatial frequency is the rate of change of the grayscale of a statistical image, reflecting the activity of the image in the spatial domain. The higher the spatial frequency, the clearer the image. Its calculation method is:

[0123] First, calculate the gradient of the image in the x and g directions, assuming that G x and G y are the gradient images in the x and y directions respectively.

[0124] Calculate the gradient magnitude image G:

[0125]

[0126] Compute the root mean square (RMS) of the gradient magnitude as a measure of spatial frequency:

[0127]

[0128] Where M and N are the width and height of the image respectively.

[0129] The standard deviation indicates the degree of dispersion of the grayscale value of the image pixels relative to the mean. The larger the standard deviation, the more dispersed the image grayscale distribution. Its calculation method is:

[0130] First, calculate the average gray value μ of the image:

[0131]

[0132] Then, the standard deviation σ is calculated using the mean gray value:

[0133]

[0134] For the fused images obtained by using different methods of Embodiment 1, Embodiment 2, Embodiment 3 and Embodiment 4, their information entropy, average gradient, spatial frequency and standard deviation evaluation values ​​are shown in Table 1 below:

[0135]

[0136] Table 1

[0137] It can be seen from the data in Table 1 that the fused image obtained by the method used in Example 1 is the best in the evaluation of information entropy and standard deviation, and is better than the weighted average method and the absolute value minimum method in terms of average gradient. The fused image obtained by the weighted average method is not satisfactory in all indicators and the fusion effect is poor; the fused image obtained by the absolute value minimum method is significantly lower than other methods in the key indicator information entropy, and significantly lower than the method in this paper in the average gradient and standard deviation indicators, so its fusion effect is poor, and the image obtained by the image fusion method based on the Prewitt operator performs better in average gradient and spatial frequency, but is still significantly lower than the method in this paper in the main indicator information entropy.

[0138] After comprehensive consideration, it is concluded that the fused image obtained by the algorithm in this paper has high contrast, contains rich information, and can provide more detailed information about the target.

[0139] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A dual-frequency disambiguation deconstruction method for military polarization images based on two-dimensional discrete waves, characterized in that: The steps include: Step S1, image acquisition: using an ordinary camera to acquire a light intensity image, by adding a polarizing plate to the ordinary camera to acquire a polarized image, and rotating the polarizing plate in front of the polarization camera to acquire polarized light intensity images in directions forming four angles of 0°, 45°, 90° and 135° with the horizontal direction; Step S2: Preliminary processing: using the formula of Stokes vector The three component images I, Q and U are obtained respectively; According to the linear polarization And polarization angle calculation formula The three component images I, Q, and U are processed and fused to obtain the linear polarization degree (D olp ) and polarization angle (A op )image; Step S3: Polarization degree fusion: In order to preserve the linear polarization degree (D olp ) and polarization angle (A op ) The information features on the image are firstly calculated using the formula Calculate D olp (x, y) and A op (x, y), and then according to the modified Toet image fusion algorithm Right D olp (x, y) and A op (x, y) is processed to obtain the polarization characteristic image D olp -A OP ; Step S4, decomposition preprocessing: In order to perform wavelet decomposition on the fused image, the intensity image and polarization feature image D are first obtained. olp -A OP Perform registration processing to make the two match in terms of avatar pixels; Step S5, wavelet decomposition and reconstruction: First, the intensity image and polarization feature image D olp -A OP After dual-frequency disparity deconstruction, the intensity image and polarization feature image D olp -A OP The low-frequency components and high-frequency components obtained by decomposition are processed in a specific way to obtain the final fused image.

2. The dual-frequency disambiguation deconstruction method for military polarization images based on two-dimensional discrete waves according to claim 1 is characterized in that: In step S2, the formula Ia of the Stokes vector (a=0°, 45°, 90°, 135°) represents the polarization image a corresponding to different polarization directions; S0 or I represents the total light intensity, that is, the sum of the light intensities in all polarization directions, which reflects the overall brightness or energy of the light beam; S1 or Q represents the intensity difference between the linear polarized light component in the horizontal direction and the linear polarized light component in the vertical direction, which reflects the difference in polarization degree of the light beam in the horizontal and vertical directions; S2 or U represents the intensity difference between the linear polarized light component at 45° and the linear polarized light component at -45° or 135°, which reveals the polarization characteristics of the light beam in the diagonal direction; S3 or V represents the intensity difference between the right-handed circularly polarized light component and the left-handed circularly polarized light component, which reflects the circular polarization component in the light beam, that is, whether the photons in the light beam rotate right or left.

3. The dual-frequency disambiguation deconstruction method for military polarization images based on two-dimensional discrete waves according to claim 1 is characterized in that: In step S2, the linear polarization degree (D olp ) represents the proportion of linear polarized light intensity in the total light intensity, and the polarization angle (A op ) represents the angle between the vibration direction of linear polarized light and the horizontal direction.

4. The dual-frequency disambiguation deconstruction method for military polarization images based on two-dimensional discrete waves according to claim 1 is characterized in that: In step S3, a, b and c are all weight factors; In the simulation experiment, the best simulation effect is achieved when a, b and c are 0.6, 0.2 and 0.2 respectively; f(x, y) is D olp With A OP The fused polarization feature image D olp -A OP .

5. The dual-frequency disambiguation deconstruction method for military polarization images based on two-dimensional discrete waves according to claim 1 is characterized in that: In step S5, the intensity image and the polarization characteristic image D olp -A OP The dual-frequency disparity deconstruction method is as follows: the intensity image and the polarization feature image D olp -A OP Perform discrete wavelet decomposition in two dimensions.

6. The dual-frequency disambiguation deconstruction method for military polarization images based on two-dimensional discrete waves according to claim 1 is characterized in that: In step S5, the intensity image and the polarization characteristic image D olp -A OP The low-frequency components and high-frequency components obtained by decomposition are processed in a specific way: for the low-frequency components, in order to retain the frequency characteristics and reflect the true characteristics of the image, the low-frequency wavelet coefficients obtained by decomposition are processed by weighted average method to make the information obtained after the fusion of information true and accurate; for the high-frequency components, in order to retain the intensity image and polarization characteristic image D olp -A OP The image structure attributes are processed by an improved fusion method based on the Sobel operator to reduce image noise and make the fused image smoother. Finally, the intensity image and polarization feature image D olp -A OP The low-frequency component and high-frequency component after fusion processing are reconstructed by two-dimensional discrete wavelet to obtain the final fused image.

7. The dual-frequency disambiguation deconstruction method for military polarization images based on two-dimensional discrete waves according to claim 6 is characterized in that: The Sobel operator includes two sets of 3x3 matrices, one for horizontal and one for vertical; The fusion method of the Sobel operator is as follows: a) performing planar convolution of the Sobel operator and the image, and the calculation equation is: b) Obtain the approximate values ​​of the horizontal and vertical brightness differences, with A representing the original image and G x and G y Respectively represent the images after horizontal and vertical edge detection, and the formulas are as follows: c) The approximate values ​​of the horizontal and vertical gradients of each pixel in the image can be expressed as Combined to calculate the magnitude of the gradient; d) Then the following formula can be used Calculate the gradient direction, and realize edge detection by calculating the gradient of the image in the x-direction and y-direction and combining the magnitude and direction information of the gradient.

8. The dual-frequency disambiguation deconstruction method for military polarization images based on two-dimensional discrete waves according to claim 1 is characterized in that: In step S5, the specific method of wavelet decomposition and reconstruction is: in the discrete case, based on the Mallat algorithm, the concept of multi-resolution analysis is used to derive a pyramid algorithm for discrete wavelet transform, and the original input signal is c0. Then the algorithm can be described as:

9. The dual-frequency disambiguation deconstruction method for military polarization images based on two-dimensional discrete waves according to claim 8, characterized in that: In the pyramid algorithm of wavelet decomposition and reconstruction, h and g are the low-pass and high-pass wavelet filters of the decomposition. For low frequency analysis signal, It is called high-frequency detail signal, J is the transformation level, ≤log2N, N is the number of samples of the original input signal; The algorithm can be extended to the case of two-dimensional image signals with separable filters. The specific method is as follows: first, each row of the image is transformed, and then each column is transformed to obtain four sub-band images representing different frequency and direction characteristics of the original image, which are denoted as LL, HL, LH, and HH respectively; Among them, LL is the low-frequency subband where the energy is concentrated, reflecting the change of grayscale; HL is the horizontal high-frequency and vertical low-frequency subbands with vertical edge information; LH is the horizontal low-frequency and vertical high-frequency subband with horizontal edge information; HH is the horizontal low-frequency vertical high-frequency subband with diagonal edge information; The LL subband can be further decomposed into four subbands to complete a multi-level two-dimensional wavelet transform.

Images