Infrared polarization multi-dimensional fusion imaging method based on nonlinear mapping

Abstract

The invention discloses an infrared polarization multi-dimensional fusion imaging method based on nonlinear mapping, and belongs to the technical field of photoelectric detection and computational imaging. Performing Stokes vector solution on the input detection data to obtain a total intensity component, a polarization degree component and a polarization angle component; performing logarithmic domain mapping and gradient calculation on the total intensity component, and performing nonlinear enhancement based on local statistical characteristics on the polarization degree component to obtain an enhanced polarization degree; in an HSV color space, constructing a brightness component according to a logarithm domain mapping result, constructing a saturation component by combining the enhanced polarization degree and gradient information, constructing a hue component according to polarization angle information, and performing threshold constraint on the hue component based on the enhanced polarization degree; and a multi-dimensional fusion image is generated through color space inverse transformation from HSV to RGB. According to the method, background clutters are effectively suppressed, the image information entropy and the average gradient are remarkably improved, and high-contrast clear imaging of a weak and small target in a complex scene is realized.

Description

Technical Field

[0001] This invention belongs to the field of photoelectric detection and computational imaging technology, specifically relating to an infrared polarization multidimensional fusion imaging method based on nonlinear mapping. Background Technology

[0002] Infrared imaging technology utilizes the difference in thermal radiation between a target and its background for imaging, offering advantages such as all-weather operation and strong smoke penetration capabilities, and is widely used in fields such as guidance, security monitoring, and assisted driving. However, due to the dual limitations of the physical properties of thermal radiation and the display quantization process, traditional infrared intensity images, while achieving high contrast through global stretching, often suffer from severe loss of local details. The fundamental reason is that the stretching process amplifies noise and compresses the subtle radiation differences between objects at the same temperature, ultimately making it difficult to effectively extract material and texture information.

[0003] Infrared polarization imaging technology further probes the polarization state information of radiation. By measuring and calculating the intensity under different polarization directions, the degree of polarization and polarization angle can be obtained. This information is closely related to the microscopic geometry and material of the object's surface, and through fusion techniques, it is possible to distinguish the structural features and material differences of the target to a certain extent.

[0004] However, current infrared polarization fusion technology faces two major challenges:

[0005] Limitations of linear processing: Traditional polarization information extraction typically employs linear stretching methods, directly stretching the degree of linear polarization. While this method can improve brightness, it also exponentially amplifies background noise, resulting in a "snowflake-like" image that severely obscures subtle texture details.

[0006] Traditional fusion methods suffer from color distortion and information conflict: Traditional weighted averaging or HSV fusion methods often struggle to properly balance the relationship between intensity and polarization information. Overemphasizing intensity information can lead to the loss of material and texture details implied by polarization; conversely, attempting to highlight polarization information can easily introduce unnatural halos, patches, or color distortions at the fusion boundary, damaging the image's realism and usability. This results in limited improvement in the information entropy (EN) and average gradient (AG) of the fused image, failing to achieve the qualitative leap from "blurred detection" to "clear perception." Therefore, effectively extracting subtle polarization details in dark areas while suppressing background noise and organically fusing them with a high-contrast intensity image is a critical problem that urgently needs to be solved in the field of infrared polarization detection. This invention aims to solve the common problems of target detail obscuring in infrared radiation imaging and the difficulty in balancing noise and texture in traditional polarization analysis, achieving high saliency perception of artificial targets in complex backgrounds. Summary of the Invention

[0007] To address the above technical problems, this invention proposes an infrared polarization multidimensional fusion imaging method based on nonlinear mapping. Stokes vector calculation is performed on the input detection data to obtain the total intensity component, polarization degree component, and polarization angle component. Logarithmic domain mapping and gradient calculation are performed on the total intensity component, and nonlinear enhancement based on local statistical features is applied to the polarization degree component to obtain the enhanced polarization degree. In the HSV color space, the luminance component is constructed using the logarithmic domain mapping result, the saturation component is constructed using the enhanced polarization degree and gradient information, and the hue component is constructed using the polarization angle information. A threshold constraint is applied to the hue component based on the enhanced polarization degree. Finally, a multidimensional fused image is generated through an inverse HSV-to-RGB color space transformation. The specific technical solution is as follows:

[0008] An infrared polarization multidimensional fusion imaging method based on nonlinear mapping includes the following steps:

[0009] Step 1: Perform Stokes vector calculation on the input detection data to obtain the total intensity component, polarization degree component, and polarization angle component;

[0010] Step 2: Perform nonlinear enhancement on the polarization component based on local statistical features to obtain the enhanced polarization. The nonlinear enhancement includes: constructing an analysis window centered on the pixel to calculate local texture activity statistics to distinguish between flat background areas and texture edge areas; constructing a dynamic adaptive gain function related to the statistics to output low gain in low-frequency background areas to suppress noise and high gain in high-frequency texture areas to enhance details; and using a sigmoid nonlinear activation function for nonlinear dynamic reconstruction.

[0011] Step 3: Perform logarithmic domain mapping and gradient calculation on the total intensity component. In the HSV color space, construct the luminance component with the logarithmic domain mapping result, construct the saturation component with the enhanced polarization degree and gradient information, construct the hue component with the polarization angle information, and apply a threshold constraint to the hue component based on the enhanced polarization degree. When the enhanced polarization degree is greater than the noise suppression threshold, use the polarization angle to generate the hue component to characterize the polarization direction feature; otherwise, set the hue component to zero.

[0012] Step 4: Generate a multidimensional fused image by performing an inverse HSV-to-RGB color space transformation on the brightness component, saturation component, and hue component.

[0013] The present invention has the following beneficial effects:

[0014] This invention adaptively extracts weak polarization textures from a noisy background using a local energy-aware nonlinear reconstruction mechanism, and achieves high-fidelity fusion of thermal radiation intensity and physical polarization properties by injecting a gradient-weighted color space into the model. The method effectively suppresses background clutter, significantly improves image information entropy and average gradient, and enables high-contrast, clear imaging of small targets in complex scenes. Attached Figure Description

[0015] Figure 1 This is a flowchart of the nonlinear dynamic range compression and HSV fusion imaging method of the present invention;

[0016] Figure 2 A schematic diagram illustrating the principle and mapping curve of the adaptive nonlinear dynamic range compression (NDRC) model;

[0017] Figure 3 A comparison image of traditional intensity, polarization degree, and multidimensional data fusion;

[0018] Figure 4 A comparison chart of metrics for different fusion methods. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other. To achieve the above objectives, this invention adopts the following technical solution.

[0020] This invention proposes an infrared polarization multidimensional fusion imaging method based on nonlinear mapping. This method solves for fundamental polarization information using Stokes theory, innovatively introduces an adaptive nonlinear mapping model to reconstruct weak texture information, and establishes a gradient-weighted inverse color space transformation model to resolve the contradiction between the "high contrast, low detail" of traditional infrared images and the "high noise, low signal-to-noise ratio" of polarization images. Figure 1 As shown, the method steps of the present invention are as follows:

[0021] Step 1: Construction and parameter solution of the full Stokes vector field.

[0022] Infrared polarization imaging units (including but not limited to focal plane, amplitude-division, or time-division detectors) are used to acquire radiation intensity data in multiple polarization directions. A Stokes vector matrix is ​​then constructed. Calculate the total radiation intensity of the scene. And the degree of linear polarization and polarization angle characterizing the surface properties of the target. The difference between horizontal and vertical polarization intensities. The difference between the intensity of 45° linear polarization and 135° linear polarization. It represents the difference in intensity between right-handed and left-handed circular polarization.

[0023] Using S0, S1, and S2, we can obtain the formulas for calculating the degree of linear polarization (DoLP) and the angle of polarization (AoP):

[0024] ;

[0025] in, To prevent correction values ​​with a denominator of zero, the original polarization image obtained at this point contains a large amount of speckle noise, and the effective texture is submerged in the low grayscale range, requiring nonlinear compression processing.

[0026] The total intensity component is obtained through Stokes parametric calculation. and linear polarization-related components , Furthermore, the linear polarization degree DoLP and polarization angle AoP are calculated and used by the subsequent intensity information processing branch and polarization information processing branch, respectively.

[0027] Step 2: Polarization enhancement based on local statistical feature perception and hyperbolic tangent dynamic compression.

[0028] A polarization information processing branch is constructed to perform NDRC nonlinear enhancement on the linear polarization degree DoLP based on local statistical feature perception, resulting in enhanced polarization degree. Meanwhile, the polarization angle AoP information is preserved as the basis for subsequent tone component construction. To address the low signal-to-noise ratio of the original polarization image, the traditional global linear stretching is abandoned, and an adaptive nonlinear mapping mechanism based on local statistical descriptors is established.

[0029] 1. Local Micro-environment Perception: An analysis window is constructed centered on an arbitrary pixel to calculate a high-order statistic representing the activity of local texture. This statistic is used to distinguish between "flat background areas" and "texture edge areas".

[0030] 2. Adaptive Gain Control Operator: Construct a dynamic adaptive gain function that is related to local statistics. The function is configured to: in the low-frequency background region (low... Output low gain to suppress noise cutoff; in high-frequency texture regions (high) Output high gain to enhance detail.

[0031] 3. Nonlinear Dynamic Reconstruction: Utilizing sigmoid nonlinear activation functions (including but not limited to hyperbolic tangent functions) A mapping model can be constructed using the sigmoid function or logistic function, and the mapping equation can be expressed as:

[0032] ;

[0033] in, Indicates the input linear polarization degree. This represents the output linear polarization degree after nonlinear dynamic reconstruction. This is a nonlinear gain coefficient used to adjust the mapping intensity; These are state control parameters used to characterize the system state, environmental disturbances, or reconfiguration conditions. For about The state modulation function is used to generate adaptive weights; It is a hyperbolic tangent sigmoid nonlinear activation function used to achieve output compression and stabilization constraints.

[0034] Step 3: Construct an HSV fusion model based on joint characterization of intensity-gradient-polarization.

[0035] An intensity information processing branch is constructed. To address the issue of excessively large dynamic range in infrared intensity images, a tone mapping operator based on a logarithmic function is used to construct the luminance component. :

[0036] ;

[0037] in, Represents pixel coordinates The luminance component of the HSV color space is used to characterize the logarithmically compressed infrared intensity information. Represents pixel coordinates The original infrared intensity value at that location, This represents the maximum value of the pixels in that image region. This indicates a normalization operation. This is a brightness compression factor used to preserve the outline details of bright targets.

[0038] To simultaneously represent physical material differences and geometric edge information in fused images, a joint injection model is constructed, including saturation components. Enhanced polarization With normalized gradient information Building together:

[0039] ;

[0040] in, The enhanced polarization degree is the output of the nonlinear compressed sensing core enhancement module in the polarization information processing branch. The intensity components of the branch are obtained from the Stokes vector solver module and input as intensity information. This is the normalized gradient term obtained after performing gradient calculation on the intensity components, used to characterize local edge and structural abrupt changes. These are the weighting coefficients.

[0041] The normalized gradient term is defined as:

[0042] ;

[0043] in, Indicates the intensity component at the pixel. gradient magnitude at that point This indicates the maximum gradient magnitude within the current image.

[0044] By constructing the saturation component as described above, the target material properties represented by polarization enhancement and the structural edge features represented by intensity gradient can be jointly mapped to the saturation component, so that the fused image can highlight polarization characteristics while maintaining clear target contours and complete local textures.

[0045] After completing the saturation component After joint injection, the polarization angle information output by the branch is further processed using polarization information. Constructing the hue components in the HSV color space Considering that the polarization angle in the low polarization region is easily affected by noise, an enhancement based on the post-polarization degree is introduced to improve the stability of tone mapping. The threshold constraint mechanism adaptively assigns values ​​to the hue components. based on Constructed, and made of enhanced polarization Threshold-based judgment control:

[0046] ;

[0047] in, Represents pixel coordinates The hue component in the HSV color space; Represents pixel coordinates The polarization angle information at that location corresponds to Figure 1 AOP information in the middle; Represents pixel coordinates Enhanced polarization degree after processing by the NDRC core enhancement module; This represents the noise suppression threshold used to suppress unstable polarization angle mapping in the low polarization region.

[0048] when At that time, it is assumed that the current pixel has relatively reliable polarization information, and the polarization angle is used. Generate hue components To characterize the polarization direction features of the target; when At that time, it was assumed that the polarization response in this region was weak or that it was subject to significant noise interference. Set to zero or a preset background value to reduce the impact of noise on tone expression and improve the stability of color mapping in the fused image.

[0049] Step 4: Generation of multidimensional fusion images based on inverse transformation from HSV to RGB color space.

[0050] Complete the hue components in the HSV color space saturation component and brightness component After construction, the three components are input. Figure 1 The color space inverse transformation module (HSV→RGB) shown is used to achieve the inverse transformation from HSV space to RGB display space, so as to obtain the final output multidimensional fused image.

[0051] For pixel coordinates Let the corresponding HSV three components be respectively , and First, define the chromaticity range factor. and intermediate variables for:

[0052] ;

[0053] ;

[0054] in, and It is the standard mathematical expression for converting the HSV color space to RGB, and its function is to convert continuous hue angles. The color is discretized into six hue interval indices, each covering 60°, corresponding to the six primary colors: red, yellow, green, cyan, blue, and magenta.

[0055] Further define intermediate variables:

[0056] ;

[0057] ;

[0058] ;

[0059] Then pixel coordinates RGB three-channel components Determined by the following piecewise function:

[0060] ;

[0061] The final fused image can be represented as:

[0062] ;

[0063] in, Represents pixel coordinates The fused image output value at the location; , and These represent the red, green, and blue channel components of the pixel in the RGB display space, respectively. , and These represent the hue component, saturation component, and brightness component obtained from the aforementioned construction, respectively. Indicates the chromaticity range number to which the hue belongs; This parameter represents the relative position of the hue within the current range; , and This is an intermediate transition variable in the inverse HSV to RGB transformation process.

[0064] Through the aforementioned inverse color space transformation, the infrared thermal radiation intensity information carried in the luminance component, the enhanced polarization texture and gradient edge information injected into the saturation component, and the polarization angle direction information represented in the hue component are uniformly mapped to the RGB display space to generate... Figure 1 The output shown is a multi-dimensional fused image. This enables the collaborative visualization of infrared intensity information, polarization texture information, and geometric structure information, improving the target recognition capability, edge sharpness, and overall color perception effect of the fused result.

[0065] In the implementation of this invention, to verify the nonlinear enhancement and multidimensional fusion effects of the proposed method, combined with Figures 2 to 4 Please provide an explanation.

[0066] Figure 2 As shown, the NDRC nonlinear mapping employed in this invention exhibits smoother dynamic response characteristics compared to the traditional truncated linear stretching method. In the low linear response region, this invention effectively suppresses background noise; in the intermediate response range, it significantly enhances the target-related signal; and in the high-brightness response range, it avoids over-enhancement or detail loss caused by traditional linear stretching through a smoothing saturation mechanism. Therefore, the nonlinear mapping method employed in this invention can balance weak target enhancement, high-brightness suppression, and overall smoothness of transition.

[0067] Figure 3 As shown, Figure 3 (a) in the image is the original intensity image, which mainly reflects the overall distribution of radiation intensity in the scene; Figure 3 (b) in the image is a polarization degree image, which can highlight the target boundary, contour and local polarization texture features; Figure 3(c) in the figure represents the fused image generated by this invention. By introducing the original intensity information, polarization enhancement information, and polarization angle information into the HSV color space for joint construction, the fused image retains the main structure of infrared intensity while further enhancing the target edges, material differences, and local texture details, thereby achieving a better overall visual perception effect.

[0068] Figure 4 As shown, the fused image generated by this invention outperforms the single input image in terms of evaluation metrics such as information entropy, average gradient, and contrast. Specifically, the increase in information entropy indicates that the fused result contains richer information; the increase in average gradient indicates that the image edges and texture details are clearer; and the increase in contrast indicates that the distinction between the target and the background is higher.

Claims

1. A method for infrared polarization multidimensional fusion imaging based on nonlinear mapping, characterized in that, Includes the following steps: Step 1: Perform Stokes vector calculation on the input detection data to obtain the total intensity component, polarization degree component, and polarization angle component; Step 2: Perform nonlinear enhancement on the polarization component based on local statistical features to obtain the enhanced polarization. The nonlinear enhancement includes: constructing an analysis window centered on the pixel to calculate local texture activity statistics to distinguish between flat background areas and texture edge areas; constructing a dynamic adaptive gain function related to the statistics to output low gain in low-frequency background areas to suppress noise and high gain in high-frequency texture areas to enhance details; and using a sigmoid nonlinear activation function for nonlinear dynamic reconstruction. Step 3: Perform logarithmic domain mapping and gradient calculation on the total intensity component. In the HSV color space, construct the luminance component with the logarithmic domain mapping result, construct the saturation component with the enhanced polarization degree and gradient information, construct the hue component with the polarization angle component, and apply a threshold constraint to the hue component based on the enhanced polarization degree. When the enhanced polarization degree is greater than the noise suppression threshold, use the polarization angle component to generate the hue component to characterize the polarization direction feature; otherwise, set the hue component to zero. Step 4: Generate a multidimensional fused image by performing an inverse HSV-to-RGB color space transformation on the brightness component, saturation component, and hue component.

2. The infrared polarization multidimensional fusion imaging method based on nonlinear mapping according to claim 1, characterized in that, The Stokes vector calculation in step 1 includes: collecting radiation intensity data in multiple polarization directions using an infrared polarization imaging unit, constructing a Stokes vector matrix, and calculating the total radiation intensity of the scene as well as the degree of linear polarization and the polarization angle.

3. The infrared polarization multidimensional fusion imaging method based on nonlinear mapping according to claim 1, characterized in that, The sigmoid nonlinear activation function mentioned in step 2 is one of the hyperbolic tangent function, sigmoid function, or logistic function.

4. The infrared polarization multidimensional fusion imaging method based on nonlinear mapping according to claim 1, characterized in that, The local texture activity statistic mentioned in step 2 is a higher-order statistic.

5. The infrared polarization multidimensional fusion imaging method based on nonlinear mapping according to claim 1, characterized in that, The logarithmic field mapping described in step 3 employs a hue mapping operator based on the logarithmic function.

6. The infrared polarization multidimensional fusion imaging method based on nonlinear mapping according to claim 1, characterized in that, The gradient calculation in step 3 is a normalized gradient term obtained after performing gradient calculation on the total intensity component, which is used to characterize local edge and structural abrupt change features.

7. The infrared polarization multidimensional fusion imaging method based on nonlinear mapping according to claim 1, characterized in that, The joint construction of the saturation components in step 3 adopts a weighted approach, and the weighting coefficients are used to adjust the fusion ratio of polarization information and gradient information.

8. The infrared polarization multidimensional fusion imaging method based on nonlinear mapping according to claim 1, characterized in that, The noise suppression threshold of the threshold constraint mentioned in step 3 is used to suppress unstable polarization angle mapping in the low polarization region.

9. The infrared polarization multidimensional fusion imaging method based on nonlinear mapping according to claim 1, characterized in that, The HSV to RGB color space inverse transformation described in step 4 includes: defining chromaticity interval factors and intermediate variables, determining the RGB three-channel components through a piecewise function, and uniformly mapping the infrared thermal radiation intensity information carried by the luminance component, the enhanced polarization texture and gradient edge information injected by the saturation component, and the polarization angle direction information represented by the hue component to the RGB display space.

10. The infrared polarization multidimensional fusion imaging method based on nonlinear mapping according to claim 2, characterized in that, Constructing the Stokes vector matrix Calculate the total radiation intensity of the scene. And the degree of linear polarization and polarization angle characterizing the surface properties of the target. The difference between horizontal and vertical polarization intensities. The difference between the intensity of 45° linear polarization and 135° linear polarization. This represents the difference in intensity between right-handed and left-handed circular polarization. Using S0, S1, and S2, the formulas for calculating the linear polarization degree DoLP and the polarization angle AoP are obtained as follows: ； in, To prevent correction values ​​where the denominator is zero.

Images