An infrared image enhancement method, system, device and storage medium based on wavelet threshold

Abstract

The application discloses an infrared image enhancement method, system, device and storage medium based on a wavelet threshold, which comprises the following steps: performing multi-scale discrete wavelet transform on an input original infrared image to separate low-frequency components and high-frequency components; the low-frequency components bear overall outlines and illumination distribution information of the image, and the high-frequency components contain edge, texture details and noise information of the image; performing improved adaptive wavelet threshold denoising processing on the high-frequency components to obtain denoised high-frequency components; performing layered enhancement processing based on weighted guided filtering and multi-scale Retinex on the low-frequency components to obtain enhanced low-frequency components; performing wavelet inverse transform reconstruction on the enhanced low-frequency components and the denoised high-frequency components to obtain a preliminary enhanced image; performing size upsampling on the preliminary enhanced image, and applying adaptive contrast restriction histogram equalization processing to output a final enhanced infrared image. The method can effectively suppress noise, retain details, adaptively enhance low-resolution infrared images, and is computationally efficient.

Description

Technical Field

[0001] This invention relates to the field of infrared imaging technology, and in particular to an infrared image enhancement method, system, device and storage medium based on wavelet threshold. Background Technology

[0002] Existing infrared image enhancement technologies mainly include the following methods, each with its own limitations: Spatial domain enhancement methods, including histogram equalization, grayscale transformation, and spatial filtering, have low computational complexity but are prone to problems such as local over-enhancement, noise amplification, and edge blurring. Transform domain enhancement methods, such as those based on wavelet transform, curvelet transform, or contourlet transform, can suppress noise to some extent, but their effectiveness in improving the inherent low contrast and detail blurring of infrared images is limited. Enhancement methods based on Retinex theory achieve enhancement by decomposing the image into illumination and reflection components and processing them separately. However, these methods often introduce visual defects such as halo artifacts, edge distortion, and staircase effects. Enhancement methods based on visual layering typically decompose the image into a basal layer and a detail layer, processing them separately. However, because they use fixed settings for the relationships between layers, they are difficult to adaptively adjust according to image content, limiting the generalization ability of enhancement performance. While enhancement methods based on set theory can improve subjective visual effects to some extent, they are prone to over-enhancement in local areas, reducing the overall naturalness of the image.

[0003] In summary, existing technologies generally suffer from the following shortcomings when processing infrared images with low illumination, low resolution, and high noise characteristics: insufficient noise suppression, severe loss of detail information, and limited contrast improvement; traditional wavelet thresholding denoising methods (such as hard thresholding and soft thresholding) are prone to signal distortion or residual noise; Retinex-type methods are prone to edge blurring and halo artifacts when correcting uneven illumination; existing layered enhancement strategies are difficult to adapt to the characteristics of infrared images in different scenarios due to the fixed division of the base layer and detail layer; in addition, traditional methods are prone to detail degradation and contrast reduction when processing small-sized infrared images (such as 32×32 pixels); some algorithms have high computational complexity, making it difficult to meet the processing efficiency requirements of embedded platforms or real-time systems. Summary of the Invention

[0004] In view of the aforementioned existing problems, this invention is proposed. Therefore, this invention provides an infrared image enhancement method, system, device, and storage medium based on wavelet thresholding to solve the problems of low contrast, blurred details, and prominent noise in low-light, low-resolution, and high-noise infrared images.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, embodiments of the present invention provide an infrared image enhancement method based on wavelet thresholding, comprising: The input raw infrared image is subjected to multi-scale discrete wavelet transform to separate low-frequency components and high-frequency components; the low-frequency components carry the overall contour and illumination distribution information of the image, while the high-frequency components contain the edge, texture details and noise information of the image. An improved adaptive wavelet threshold denoising process is performed on the high-frequency components to obtain denoised high-frequency components; The low-frequency components are subjected to hierarchical enhancement processing based on weighted guided filtering and multi-scale Retinex to obtain enhanced low-frequency components; The enhanced low-frequency component and the denoised high-frequency component are reconstructed by inverse wavelet transform to obtain a preliminary enhanced image. The initial enhanced image is upsampled and subjected to adaptive contrast-limited histogram equalization to output the final enhanced infrared image.

[0006] As a preferred embodiment of the infrared image enhancement method based on wavelet thresholding described in this invention, the step of performing multi-scale discrete wavelet transform on the input original infrared image to separate low-frequency components and high-frequency components includes: the multi-scale discrete wavelet transform is performed using the Daubechies 4 wavelet basis for three-level decomposition; the low-frequency component is the low-frequency approximate sub-band obtained by the third-level decomposition, and the high-frequency component is the set of horizontal, vertical, and diagonal high-frequency detail sub-bands obtained by the first to third-level decompositions.

[0007] As a preferred embodiment of the infrared image enhancement method based on wavelet thresholding described in this invention, wherein: improved adaptive wavelet thresholding denoising processing is performed on the high-frequency components to obtain denoised high-frequency components, including: The initial values ​​of the noise standard deviation for each high-frequency subband were estimated using the median absolute deviation method. An adaptive threshold is calculated based on the initial noise standard deviation, the reaction balance parameter, and the structure adjustment parameter. The reaction balance parameter is the ratio of the number of coefficients in the current high-frequency subband whose absolute value is less than the initial general threshold to the total number of coefficients in that subband. The structure adjustment parameter is an empirical constant dynamically adjusted based on the global contrast of the image.

[0008] As a preferred embodiment of the infrared image enhancement method based on wavelet thresholding described in this invention, the method further includes: the threshold function used in the improved adaptive wavelet thresholding denoising process is a continuously differentiable nonlinear mapping function; the threshold function achieves continuous first derivative at the point where the absolute value of the high-frequency subband coefficients is equal to the adaptive threshold to eliminate the pseudo-Gibbs effect, and when the absolute value of the coefficients is greater than the adaptive threshold, the function output value approaches the soft thresholding shrinkage result.

[0009] As a preferred embodiment of the infrared image enhancement method based on wavelet threshold described in this invention, wherein: the low-frequency component is subjected to hierarchical enhancement processing based on weighted guided filtering and multi-scale Retinex to obtain enhanced low-frequency components, including: when the weighted guided filtering decomposes the low-frequency component, the weighting factor is dynamically constructed using the variance and regularization parameters of the guided image within the local window; The configuration of the weighting factor enables the filtering process to apply strong smoothing to flat areas of the image to separate the illumination distribution, and to apply weaker filtering to structural edge areas to preserve detailed contours. The base layer obtained by weighted guided filtering is subjected to multi-scale Retinex processing using Gaussian surround functions of three scales for illumination estimation and correction, and the corrected base layer is restored by exponential transformation; wherein, the multi-scale Retinex processing uses the output of weighted guided filtering as guided input to suppress halo artifacts.

[0010] As a preferred embodiment of the infrared image enhancement method based on wavelet threshold described in this invention, the method of reconstructing the enhanced low-frequency component and the denoised high-frequency component by performing inverse wavelet transform to obtain a preliminary enhanced image includes: applying an adaptive gain to the detail layer obtained by weighted guided filtering, wherein the adaptive gain is jointly determined by the base gain coefficient, the edge modulation factor and the gradient magnitude of the low-frequency component. The corrected base layer and the detail layer enhanced by adaptive gain are added at the pixel level and fused to obtain the enhanced low-frequency component.

[0011] As a preferred embodiment of the wavelet threshold-based infrared image enhancement method of the present invention, the method includes: upsampling the size of the preliminary enhanced image and applying adaptive contrast-limited histogram equalization to output the final enhanced infrared image, comprising: A bicubic interpolation algorithm is used to upsample the original low-resolution infrared image to a preset target display resolution; the adaptive contrast-limited histogram equalization process is applied to the upsampled image in a slightly enhanced mode to optimize the global contrast distribution and suppress local pixel oversaturation.

[0012] Secondly, the present invention provides an infrared image enhancement system based on wavelet thresholding, comprising: The multi-scale decomposition module is used to perform multi-scale discrete wavelet transform on the input raw infrared image to separate low-frequency components and high-frequency components; the low-frequency components carry the overall contour and illumination distribution information of the image, and the high-frequency components contain the edge, texture details and noise information of the image; A high-frequency denoising module is used to perform improved adaptive wavelet threshold denoising processing on the high-frequency components to obtain denoised high-frequency components. The low-frequency enhancement module is used to perform hierarchical enhancement processing based on weighted guided filtering and multi-scale Retinex on the low-frequency components to obtain enhanced low-frequency components. The image reconstruction module is used to reconstruct the enhanced low-frequency component and the denoised high-frequency component by performing inverse wavelet transform to obtain a preliminary enhanced image. The post-processing optimization module is used to upsample the size of the preliminary enhanced image and apply adaptive contrast-limited histogram equalization to output the final enhanced infrared image.

[0013] Thirdly, the present invention provides an electronic device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the wavelet threshold-based infrared image enhancement method.

[0014] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the wavelet threshold-based infrared image enhancement method.

[0015] Compared with existing technologies, the advantages of this invention are as follows: By employing an improved adaptive wavelet threshold function, this invention can effectively remove Gaussian noise and impulse noise in infrared images, resulting in a significant improvement in the peak signal-to-noise ratio and structural similarity index of the processed image. By combining weighted guided filtering with multi-scale Retinex, the image detail layer is enhanced, thereby better preserving the edge sharpness and texture information of the image. This invention can adaptively adjust the wavelet threshold and filtering parameters according to the specific content of the input image, thus effectively adapting to the infrared image enhancement needs in different scenarios. It has a significant enhancement effect on small-sized, low-resolution infrared images, effectively improving their visual quality and enhancing the reliability of subsequent feature extraction. The overall structure of the algorithm of this invention is clear, the computational complexity is moderate, it is easy to deploy on embedded platforms, and it is suitable for real-time infrared imaging systems. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a schematic flowchart of an infrared image enhancement method based on wavelet threshold according to an embodiment of the present invention; Figure 2 The image shows the evaluation results of several algorithms in the first experiment of an infrared image enhancement method based on wavelet thresholding according to an embodiment of the present invention. Figure 3 This is a thermal imaging image and a sub-band image schematic diagram of an infrared image enhancement method based on wavelet thresholding according to an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating the test results of various blurring techniques used in an embodiment of the present invention to describe an infrared image enhancement method based on wavelet thresholding. Figure 5 This is a schematic diagram comparing the processing effects of an infrared image enhancement method based on wavelet thresholding according to an embodiment of the present invention on the same real infrared thermal imaging data. Detailed Implementation

[0017] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0018] Example 1, referring to Figure 1 This is one embodiment of the present invention, which provides an infrared image enhancement method based on wavelet thresholding, comprising: S100: Performs multi-scale discrete wavelet transform on the input raw infrared image to separate low-frequency components and high-frequency components; the low-frequency components carry the overall contour and illumination distribution information of the image, while the high-frequency components contain the edge, texture details and noise information of the image. S200: Perform improved adaptive wavelet threshold denoising on the high-frequency components to obtain denoised high-frequency components; S300: Performs hierarchical enhancement processing based on weighted guided filtering and multi-scale Retinex on the low-frequency components to obtain enhanced low-frequency components; S400: The enhanced low-frequency components and the denoised high-frequency components are reconstructed by inverse wavelet transform to obtain the preliminary enhanced image; S500: Upsamples the initial enhanced image and applies adaptive contrast-limited histogram equalization to output the final enhanced infrared image.

[0019] It should be noted that traditional wavelet thresholding methods are prone to signal distortion or noise residue during denoising. Retinex-type methods tend to cause edge blurring and halo artifacts when removing uneven illumination. Existing layered enhancement methods have a relatively fixed division between the base layer and detail layer, making it difficult to adapt to the characteristics of infrared images in different scenarios.

[0020] Traditional methods for enhancing small-sized infrared images often suffer from detail loss and decreased contrast. Some methods also incur high computational costs for real-time enhancement, making them unsuitable for embedded or real-time systems. This invention aims to systematically address the problems of low contrast, blurred details, and prominent noise in low-light, low-resolution, and high-noise infrared images by constructing a multi-layered, adaptive processing framework that sequentially performs noise suppression, illumination correction, and detail enhancement.

[0021] In this embodiment of the invention, step S100 involves performing a multi-scale discrete wavelet transform on the input raw infrared image to separate low-frequency components and high-frequency components. This includes: the multi-scale discrete wavelet transform is performed using the Daubechies 4 wavelet basis for a 3-level decomposition; the low-frequency component is the low-frequency approximate subband obtained by the 3rd level decomposition, and the high-frequency component is the set of horizontal, vertical, and diagonal high-frequency detail subbands obtained by the 1st to 3rd level decomposition.

[0022] In one feasible approach, the input raw infrared image... Discrete wavelet transform is performed. In this embodiment of the invention, the Daubechies 4 wavelet basis is preferably used for three-level decomposition. This decomposition process decouples the enhancement problem into illumination processing of low-frequency components and noise suppression of high-frequency components. Each level of decomposition divides the image into four sub-bands, where: The low-frequency approximation subband (LL) carries the overall contour of the image and the gradually varying illumination information.

[0023] The horizontal high-frequency detail subband (LH) mainly contains the edges and details in the vertical direction.

[0024] The vertical high-frequency detail subband (HL) mainly contains the edges and details in the horizontal direction.

[0025] The diagonal high-frequency detail subband (HH) mainly contains details in the diagonal direction and a lot of noise.

[0026] After multi-level decomposition, the final low-frequency sub-bands are collectively referred to as low-frequency components, which contain the main structure and illumination distribution of the image; all levels of high-frequency sub-bands (LH, HL, HH) are collectively referred to as high-frequency components, which contain the image's edges, texture details, and most of the noise.

[0027] It should be noted that traditional hard thresholding functions are prone to causing pseudo-Gibbs effects, while soft thresholding functions can lead to overly smoothed edges. This invention provides a continuously differentiable adaptive thresholding function to achieve an optimal balance between noise suppression and detail preservation.

[0028] In this embodiment of the invention, step S200 involves performing improved adaptive wavelet threshold denoising on the high-frequency components to obtain denoised high-frequency components, including: The initial values ​​of the noise standard deviation for each high-frequency subband were estimated using the median absolute deviation method. The adaptive threshold is calculated based on the initial value of the noise standard deviation, the reaction balance parameter, and the structure adjustment parameter. The reaction balance parameter is the ratio of the number of coefficients in the current high-frequency subband whose absolute value is less than the initial general threshold to the total number of coefficients in that subband. The structure adjustment parameter is an empirical constant that is dynamically adjusted based on the global contrast of the image.

[0029] In this embodiment of the invention, step S200 further includes: the threshold function used in the improved adaptive wavelet threshold denoising process is a continuously differentiable nonlinear mapping function; the threshold function achieves continuous first derivative at the point where the absolute value of the high-frequency subband coefficients is equal to the adaptive threshold to eliminate the pseudo-Gibbs effect, and when the absolute value of the coefficients is greater than the adaptive threshold, the function output value approaches the soft threshold shrinkage result.

[0030] In one feasible approach, for each high-frequency subband of the j-th level decomposition, the initial value of the noise standard deviation of that subband is estimated using a robust MedianAbsolute Deviation method:

[0031] in, denoted as the high-frequency subband of the j-th layer decomposition, and k represents the horizontal, vertical, and diagonal high-frequency detail subbands.

[0032] Furthermore, key adjustment parameters are introduced: the reaction equilibrium parameter S and the structure adjustment parameter m. The threshold of the k-th subband in the j-th layer is also discussed. The calculation is as follows:

[0033] in, It is the total number of coefficients in the j-th subband. , This is the number of coefficients in the subband whose absolute value is less than the current iteration threshold (initially a general threshold). S reflects the proportion of "suspected noise" coefficients in the subband, and m is an empirical constant dynamically adjusted based on the global contrast of the image, typically between 0.5 and 1.5, used to adjust the threshold level in scenes with strong noise and weak detail.

[0034] Furthermore, define a new threshold processing function. Where w are wavelet coefficients:

[0035] It should be noted that this function is continuously differentiable at |w| = T, overcoming the discontinuity problem of hard thresholding. When |w| is much greater than T, the function behavior approaches soft thresholding (wT), effectively preserving the signal; when |w| is close to T, the function provides a smooth transition, reducing artifact generation. The high-frequency subband processed by this function is denoted as […]. The noise is significantly suppressed, while the important edge coefficients are preserved.

[0036] It should be noted that the low-frequency components include uneven illumination. In this embodiment of the invention, a weighted guided filter is used to decompose the low-frequency components into a base layer and a detail layer, and then multi-scale Retinex illumination correction is performed on the base layer.

[0037] In this embodiment of the invention, step S300 performs a layered enhancement process based on weighted guided filtering and multi-scale Retinex on the low-frequency components to obtain enhanced low-frequency components, including: when the weighted guided filtering decomposes the low-frequency components, it dynamically constructs weight factors based on the variance and regularization parameters of the guided image within the local window. The configuration of the weighting factor enables the filtering process to apply strong smoothing to flat areas of the image to separate the illumination distribution, and to apply weaker filtering to structural edge areas to preserve detailed contours. The base layer obtained by weighted guided filtering is processed by multi-scale Retinex using Gaussian wrap functions of three scales to perform illumination estimation and correction, and the corrected base layer is restored by exponential transformation. The multi-scale Retinex processing uses the output of weighted guided filtering as the guiding input to suppress halo artifacts.

[0038] In one feasible approach, the low-frequency components are used as the guide image G and the input image I (here, the same image), and decomposed by a weighted guided filter:

[0039] Here, B(x,y) is the base layer of the output (smoothing lighting layer). It is a local window centered at pixel k. Coefficients , The solution is found by minimizing the following weighted cost function:

[0040] Furthermore, weighting factors Determined by the edge information of the guiding image G:

[0041] in, It is a window The variance of G is given by λ, which is the regularization parameter, and N is the total number of image windows. This weighting results in stronger filtering in flat regions (smoothing illumination) and weaker filtering in edge regions (preserving details). The detail layer D(x,y) is then calculated using the residuals. D(x,y)=LFS(x,y)-B(x,y).

[0042] Furthermore, MSR processing is applied to the base layer B(x,y) to eliminate uneven illumination. A Gaussian wrap function is used. (x, y), select three scales (small, medium, large):

[0043] Where * represents convolution, The weights for each scale are usually equal, with 1 / 3 being the average weight.

[0044] It should be noted that the output of the weighted guided filter is used as a guide to improve Gaussian wrapping. That is, the result of the weighted guided filter is used to simulate lighting estimates that better reflect the actual scene, thus avoiding edge halos caused by traditional Gaussian wrapping. Corrected base layer Recovery via exponential transformation: .

[0045] In this embodiment of the invention, step S400, which involves performing wavelet inverse transform reconstruction on the enhanced low-frequency component and the denoised high-frequency component to obtain a preliminary enhanced image, includes: applying an adaptive gain to the detail layer obtained by weighted guided filtering. The adaptive gain is determined by the base gain coefficient, the edge modulation factor, and the gradient magnitude of the low-frequency component. The corrected base layer and the detail layer enhanced by adaptive gain are added at the pixel level and fused to obtain the enhanced low-frequency component.

[0046] Furthermore, adaptive gain is applied to the detail layer D(x,y) to enhance the texture:

[0047] Where α is the fundamental gain coefficient, β is the edge modulation factor, and ||LFS| is the gradient magnitude of LFS, used to provide higher gain in the edge region. The final enhanced low-frequency component... It is formed by merging the enhanced base layer and detail layer:

[0048] In this embodiment of the invention, step S500 involves upsampling the size of the preliminary enhanced image and applying adaptive contrast-limited histogram equalization to output the final enhanced infrared image, including: A bicubic interpolation algorithm is used to upsample the original low-resolution infrared image to a preset target display resolution. Adaptive contrast-limited histogram equalization is applied to the upsampled image in a slightly enhanced mode to optimize the global contrast distribution and suppress local pixel oversaturation.

[0049] Specifically, the processed low-frequency components With the denoised high-frequency components at each level Perform inverse wavelet transform to reconstruct the preliminary enhanced image. For extremely low resolution, such as a raw input of 32×32 pixels, bicubic interpolation upsampling can be used after reconstruction to the target display size, such as 320×320. A slight adaptive contrast-limiting histogram equalization is then applied to the upsampled image to further optimize the overall contrast and avoid oversaturation in local areas, resulting in the final enhanced image. .

[0050] The above is an illustrative scheme of an infrared image enhancement method based on wavelet thresholding according to this embodiment. It should be noted that the technical solution of this infrared image enhancement system based on wavelet thresholding belongs to the same concept as the above-described infrared image enhancement method based on wavelet thresholding. Details not described in detail in the technical solution of the infrared image enhancement system based on wavelet thresholding in this embodiment can be found in the description of the above-described infrared image enhancement method based on wavelet thresholding.

[0051] This embodiment provides an infrared image enhancement system based on wavelet thresholding, comprising: The multi-scale decomposition module is used to perform multi-scale discrete wavelet transform on the input raw infrared image to separate low-frequency components and high-frequency components; the low-frequency components carry the overall contour and illumination distribution information of the image, while the high-frequency components contain the edge, texture details and noise information of the image. The high-frequency denoising module is used to perform improved adaptive wavelet threshold denoising on the high-frequency components to obtain denoised high-frequency components. The low-frequency enhancement module is used to perform hierarchical enhancement processing on low-frequency components based on weighted guided filtering and multi-scale Retinex to obtain enhanced low-frequency components. The image reconstruction module is used to reconstruct the enhanced low-frequency components and the denoised high-frequency components by inverse wavelet transform to obtain a preliminary enhanced image. The post-processing optimization module is used to upsample the size of the preliminary enhanced image and apply adaptive contrast-limited histogram equalization to output the final enhanced infrared image.

[0052] It should be noted that the entire algorithm can be integrated into an embedded infrared imaging system. The system workflow is as follows: sensor acquisition (low-resolution infrared data) → wavelet decomposition → HFS adaptive threshold denoising → LFS WGIF decomposition and MSR enhancement → wavelet reconstruction → upsampling and post-processing → output of high-quality enhanced image. This workflow ensures end-to-end optimization from noise suppression and illumination correction to detail enhancement, making it particularly suitable for processing image data from miniature thermal sensors.

[0053] This embodiment also provides an electronic device applicable to infrared image enhancement methods based on wavelet thresholds, including: The system includes a memory and a processor. The memory stores computer-executable instructions, and the processor executes these instructions to implement the wavelet threshold-based infrared image enhancement method as described in the above embodiments.

[0054] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the wavelet threshold-based infrared image enhancement method as proposed in the above embodiments.

[0055] The storage medium proposed in this embodiment and the infrared image enhancement method based on wavelet threshold proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0056] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0057] Example 2, refer to Figure 2 This is one embodiment of the present invention. Unlike the first embodiment, this embodiment verifies the beneficial effects of the present invention through comparative experiments.

[0058] This embodiment uses infrared images from a laboratory dataset, as well as infrared thermal imaging images acquired in the field using a thermal scanning device. The test images cover complex scenes containing details such as leaves, shrubs, and building entrances; To comprehensively evaluate the performance of the method of the present invention, this embodiment selects five existing mainstream infrared image enhancement algorithms as comparison methods, and processes them with the method of the present invention under the same parameter conditions. The comparison methods include brightness-preserving dual histogram equalization, multi-scale Retinex, bilateral filtering and detail-preserving processing, and weighted guided filtering.

[0059] Figure 2 The results of the enhancement processing of the same test image by the different algorithms mentioned above are shown. Figure 2 (a) is the result of subjective evaluation. Figure 2 (b) Results showed overexposure in the foliage and shrub areas (see the green box area). Figure 2 (c) The results show that the details in the areas of leaves and green plants are rather dim. Figure 2 (d) The results show that the details are relatively clear, but a lot of noise is immediately introduced; Figure 2 (e) An excessively high base layer fusion ratio leads to more severe deformation; Figure 2 (f) The method of the present invention significantly improves image contrast, the wavelet threshold effectively eliminates noise, and the details and contours are clearly distinguishable.

[0060] Example 3, referring to Figure 3 This is one embodiment of the present invention. Unlike the first embodiment, this embodiment details the specific implementation of multi-scale decomposition of infrared images using wavelet transform.

[0061] In this embodiment, the Daubechies wavelet basis is used to perform multi-resolution decomposition on the infrared thermal imaging image. The input infrared image is decomposed into low-frequency sub-bands and high-frequency sub-bands through wavelet transform. The high-frequency sub-band is further divided into horizontal sub-bands, vertical sub-bands, and diagonal sub-bands at different decomposition scales. Figure 3 As shown, the top left corner displays the original thermal imaging image (or approximate sub-band), while the remaining areas show the high-frequency detail sub-band images corresponding to each decomposition level, intuitively demonstrating the multi-scale frequency band separation effect of wavelet transform on infrared images.

[0062] Example 4, refer to Figure 4 This is one embodiment of the present invention. Unlike the first embodiment, this embodiment verifies the technical effect of the improved wavelet threshold function proposed in the present invention in infrared image denoising through comparative testing.

[0063] In this embodiment, a test infrared image is selected. Figure 4(a) As input sample. To simulate actual imaging noise, Gaussian white noise with a mean of zero and a variance of 0.01 is artificially added to the image. The noisy image is then processed using hard thresholding, soft thresholding, and the threshold function method improved in this invention, respectively: peak signal-to-noise ratio and structural similarity are used as objective evaluation indicators.

[0064] To facilitate a clearer analysis of the noise reduction effect, in Figure 4 Enlarged images of the "license plate number" area were inserted in the upper left corner of (b), (c), and (d); Figure 4 (b) Hard thresholding method: Despite the use of a strict hard thresholding noise reduction method, the details of the "license plate number" still contain a small amount of noise residue; Figure 4 (c) When the traditional soft thresholding method is used, the outer contour of the license plate number is distorted. Figure 4 (d) After processing with the improved threshold function method, the appearance depth is clear, the edge contour is easy to identify, and the gradient artifacts of the image can be more effectively suppressed and eliminated.

[0065] Additional objective evaluation: White Gaussian noise of varying intensities was injected into the image, and hard thresholding, soft thresholding, and an improved thresholding function were applied for processing. The peak signal-to-noise ratio and structural similarity values ​​of each method were calculated. The results show that the improved thresholding method has better denoising efficiency than the two traditional methods mentioned above.

[0066] Example 5, refer to Figure 5 This is one embodiment of the present invention. Unlike the first embodiment, this embodiment verifies the enhancement effect of the method of the present invention on extremely low resolution real infrared thermal imaging data through comparative testing.

[0067] The experimental platform in this embodiment is an infrared thermal imaging analysis subsystem. The second set of experiments is a real-world application scenario verification. The subsystem receives infrared thermal radiation from surrounding objects and converts it into raw images. The test scenario includes the facial area of ​​a person.

[0068] The imaging device has a resolution of 32×32 pixels; the preprocessing method uses the nearest neighbor interpolation algorithm to uniformly enlarge the original thermal image to 320×320 pixels. Figure 5 The paper presents a comparison of the processing effects of the method of this invention with various existing algorithms on the same real infrared thermal imaging data. It can be seen that after the nearest neighbor interpolation magnification, the brightness of the facial image is improved, but the details are not clear, and the outline of the person's face and the boundary outline of the "table lamp" in the background are relatively blurry.

[0069] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An infrared image enhancement method based on wavelet thresholding, characterized in that, include: Multi-scale discrete wavelet transform is performed on the input raw infrared image to separate low-frequency and high-frequency components. The low-frequency component carries the overall contour and illumination distribution information of the image, while the high-frequency component includes the image's edge, texture details, and noise information. An improved adaptive wavelet threshold denoising process is performed on the high-frequency components to obtain denoised high-frequency components; The low-frequency components are subjected to hierarchical enhancement processing based on weighted guided filtering and multi-scale Retinex to obtain enhanced low-frequency components; The enhanced low-frequency component and the denoised high-frequency component are reconstructed by inverse wavelet transform to obtain a preliminary enhanced image. The initial enhanced image is upsampled and subjected to adaptive contrast-limited histogram equalization to output the final enhanced infrared image.

2. The infrared image enhancement method based on wavelet threshold as described in claim 1, characterized in that, The step of performing a multi-scale discrete wavelet transform on the input raw infrared image to separate low-frequency and high-frequency components includes: the multi-scale discrete wavelet transform is performed using the Daubechies 4 wavelet basis for a 3-level decomposition; the low-frequency component is the low-frequency approximate sub-band obtained by the 3rd level decomposition, and the high-frequency component is the set of horizontal, vertical, and diagonal high-frequency detail sub-bands obtained by the 1st to 3rd level decomposition.

3. The infrared image enhancement method based on wavelet threshold as described in claim 2, characterized in that, The improved adaptive wavelet threshold denoising process is performed on the high-frequency components to obtain the denoised high-frequency components, including: The initial values ​​of the noise standard deviation for each high-frequency subband were estimated using the median absolute deviation method. An adaptive threshold is calculated based on the initial noise standard deviation, the reaction balance parameter, and the structure adjustment parameter. The reaction balance parameter is the ratio of the number of coefficients in the current high-frequency subband whose absolute value is less than the initial general threshold to the total number of coefficients in that subband. The structure adjustment parameter is an empirical constant dynamically adjusted based on the global contrast of the image.

4. The infrared image enhancement method based on wavelet threshold as described in claim 3, characterized in that, Also includes: The threshold function used in the improved adaptive wavelet thresholding denoising process is a continuously differentiable nonlinear mapping function. The threshold function achieves first-order derivative continuity when the absolute value of the high-frequency subband coefficients equals the adaptive threshold to eliminate the pseudo-Gibbs effect, and when the absolute value of the coefficients is greater than the adaptive threshold, the function output value approaches the soft threshold contraction result.

5. The infrared image enhancement method based on wavelet threshold as described in claim 4, characterized in that, The low-frequency components are subjected to hierarchical enhancement processing based on weighted guided filtering and multi-scale Retinex to obtain enhanced low-frequency components, including: when the weighted guided filtering decomposes the low-frequency components, the weighting factors are dynamically constructed using the variance and regularization parameters of the guided image within a local window. The configuration of the weighting factor enables the filtering process to apply strong smoothing to flat areas of the image to separate the illumination distribution, and to apply weaker filtering to structural edge areas to preserve detailed contours. The base layer obtained by weighted guided filtering is subjected to multi-scale Retinex processing using Gaussian surround functions of three scales for illumination estimation and correction, and the corrected base layer is restored by exponential transformation; wherein, the multi-scale Retinex processing uses the output of weighted guided filtering as guided input to suppress halo artifacts.

6. The infrared image enhancement method based on wavelet threshold as described in claim 5, characterized in that, The process of reconstructing the enhanced low-frequency component and the denoised high-frequency component by performing inverse wavelet transform to obtain the preliminary enhanced image includes: applying an adaptive gain to the detail layer obtained by weighted guided filtering, wherein the adaptive gain is jointly determined by the base gain coefficient, the edge modulation factor and the gradient magnitude of the low-frequency component; The corrected base layer and the adaptively enhanced detail layer are added pixel-by-pixel and fused to obtain the enhanced low-frequency component.

7. The infrared image enhancement method based on wavelet threshold as described in claim 6, characterized in that, The initial enhanced image is upsampled and subjected to adaptive contrast-limited histogram equalization to output the final enhanced infrared image, including: A bicubic interpolation algorithm is used to upsample the original low-resolution infrared image to a preset target display resolution; the adaptive contrast-limited histogram equalization process is applied to the upsampled image in a slightly enhanced mode to optimize the global contrast distribution and suppress local pixel oversaturation.

8. An infrared image enhancement system based on wavelet thresholding, applied to the method described in any one of claims 1-7, characterized in that, include: The multi-scale decomposition module is used to perform multi-scale discrete wavelet transform on the input raw infrared image to separate low-frequency components from high-frequency components. The low-frequency component carries the overall contour and illumination distribution information of the image, while the high-frequency component includes the image's edge, texture details, and noise information. A high-frequency denoising module is used to perform improved adaptive wavelet threshold denoising processing on the high-frequency components to obtain denoised high-frequency components. The low-frequency enhancement module is used to perform hierarchical enhancement processing based on weighted guided filtering and multi-scale Retinex on the low-frequency components to obtain enhanced low-frequency components; The image reconstruction module is used to reconstruct the enhanced low-frequency component and the denoised high-frequency component by performing inverse wavelet transform to obtain a preliminary enhanced image. The post-processing optimization module is used to upsample the size of the preliminary enhanced image and apply adaptive contrast-limited histogram equalization to output the final enhanced infrared image.

9. An electronic device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the wavelet threshold-based infrared image enhancement method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the wavelet threshold-based infrared image enhancement method according to any one of claims 1 to 7.

Images