Adaptive processing method and system for high dynamic infrared image of complex scene

Abstract

The application discloses a high-dynamic infrared image adaptive processing method and system for a complex scene, which comprises the following steps: performing logarithmic domain transformation on high-bit-depth thermal infrared images pixel by pixel and counting the histogram of each gray level; screening effective gray levels according to the frequency support degree, spatial aggregation degree and relative intensity of the gray levels to obtain an effective histogram; performing logarithmic weighting and constructing a gray distribution lookup table according to a cumulative distribution function to adjust the gray output expression proportion of a target and a background; constructing a detail enhancement lookup table and a brightness weight lookup table, performing nonlinear stretching on normalized gray values, and determining the detail weight according to the normalized brightness values; performing table lookup fusion based on the gray distribution lookup table, the detail enhancement lookup table and the brightness weight lookup table; and linearly normalizing the fusion result and mapping it to a limited bit-width output gray range; and the application can maintain the distinguishability of target edges and internal textures in the 8-bit output image on the basis of effective content expression and target-background relationship expression.

Description

Technical Field

[0001] This invention relates to the field of infrared image processing technology, and more specifically to a method and system for adaptive processing of high dynamic infrared images in complex scenes. Background Technology

[0002] Thermal infrared images can reflect the difference in thermal radiation between a target and the background, and still have good target perception capabilities under low light, occlusion, and complex background conditions. With the development of low-altitude economy and intelligent sensing technology, they are widely used in scenarios such as UAV inspection, security monitoring, emergency rescue, night monitoring, and reconnaissance in complex environments. However, in practical applications, thermal infrared images are often affected by backlit sky backgrounds, surface thermal reflection, building shadows, complex thermal fields at night, and local firelight, resulting in problems such as local overbrightness, obscuring of weak target details, and overall dynamic range imbalance. At the same time, due to detector non-uniformity, readout noise, and dead pixels, there are also randomly distributed isolated gray-scale points in the images. These factors directly affect the accuracy of subsequent target detection and scene perception.

[0003] Thermal infrared sensor outputs typically have a high bit depth, such as 14 bits, while back-end processing links and display devices mostly use 8 bits or other limited bit widths. Because the output grayscale space is limited when converting high bit depth data to 8-bit or other limited bit width images, different grayscale information in the image needs to be redistributed within the limited space. Therefore, the weak target edges, dark area textures, and local highlight differences that can be distinguished in a 14-bit image cannot be preserved in equal amounts in an 8-bit output. However, if linear compression is used directly, the bright area levels and dark area details are easily lost, and the edges and textures of weak targets are difficult to preserve stably.

[0004] Temperature information in complex scenes has obvious high dynamic characteristics. Typical scenes include low-temperature targets in low-temperature environments, high-temperature targets such as explosions and flames, and scenes where low-temperature targets, medium-temperature backgrounds, and high-temperature regions coexist in the same field of view. The thermal radiation difference between low-temperature environments or low-temperature targets and the surrounding background is small, resulting in weak gray-level differences in the corresponding images. Target edges and textures are easily mixed with the background gray level. The corresponding areas of high-temperature targets such as explosions and flames may exhibit localized highlights or even saturation, which can easily occupy a large amount of output gray-level space and affect the expression of surrounding information.

[0005] When the temperature range within the same scene is large, the image grayscale information corresponding to different temperature states is distributed over a wide range of input grayscale. Low-temperature targets, medium-temperature backgrounds, high-temperature targets, and local bright areas need to be mapped to a limited range of output grayscale. Without adaptive processing for the grayscale distribution of the scene and the relationship between the target and the background, it is difficult to stably represent the weak grayscale differences of low-temperature targets, the necessary layers of the background, and the local structure of high-temperature areas at the same time.

[0006] Existing histogram equalization methods allocate the output grayscale space according to the frequency of occurrence of each grayscale level. Large flat backgrounds correspond to a large number of grayscale pixels, which tend to occupy a lot of output space; dark and weak targets correspond to a small number of grayscale pixels, which tend to be compressed after mapping, making it difficult to identify weak targets.

[0007] Other improved methods, such as platform histograms and dual platform histograms, limit the counting of high-frequency gray levels by setting a platform threshold to alleviate the background dominance problem. However, the platform threshold needs to be determined adaptively according to the scene, and the parameter selection is relatively complex. Therefore, invalid gray levels caused by detector noise, bad pixels, and local abnormal brightness may still participate in gray level allocation and be amplified during the stretching process, affecting the quality of the output image.

[0008] To balance denoising and enhancement, some solutions first use methods such as bilateral filtering, guided filtering, or nonlocal means to filter pixels one by one in the image domain, and then perform histogram enhancement. This type of image domain filtering requires neighborhood operations, which involves a large amount of computation and frequent storage access. Furthermore, the denoising and enhancement parameters are difficult to coordinate, making it unsuitable for real-time implementation on edge computing hardware with limited computing power and power consumption.

[0009] Therefore, how to adaptively reconstruct and allocate the limited output grayscale space based on the image grayscale distribution, spatial relationship, and grayscale relationship between the target and the background corresponding to different temperature states during the conversion of high bit depth dynamic infrared images in complex scenes to limited bit width, improve the presentation effect of low temperature targets, high temperature targets, and background information in the output image, reduce the occupation of the limited output grayscale space by invalid interference grayscale and large area background, maintain the recognizability of target edges and internal textures, and take into account the real-time implementation requirements of edge computing hardware, is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0010] In view of the above problems, this invention proposes an adaptive processing method and system for high dynamic range infrared images in complex scenes to overcome or at least partially solve the above problems. Addressing the grayscale information representation requirements in the conversion process from high dynamic range infrared images in complex scenes to images with limited bit width, the method filters effective grayscale information and reduces interference grayscale occupancy based on the relationship between effective and invalid content in the image. It reconstructs the output grayscale allocation relationship based on the grayscale relationship between the target and the background to improve the distinguishability between the target and the background. Furthermore, it enhances details and imposes weight constraints based on the target's own edge, texture, and dark level representation requirements, ultimately outputting a grayscale map to complete the conversion from high dynamic range infrared images in complex scenes to images with limited bit width.

[0011] To achieve the above objectives, the present invention adopts the following technical solution:

[0012] A high dynamic range infrared image adaptive processing method for complex scenes includes: S1. Perform a logarithmic domain transformation on each pixel of the input high-bit deep thermal infrared image to obtain a logarithmic domain image, and then normalize it before calculating the histogram of each gray level. S2. Based on the frequency support, spatial clustering and relative intensity of each gray level in the gray-level histogram, the effective gray levels are selected to obtain the effective histogram, thereby reducing the occupancy of the output gray-level space by interfering gray levels; S3. Log-weight the effective histogram and construct a gray-level allocation lookup table based on the cumulative distribution function of the log-weighted histogram to adjust the output ratio of target-related gray levels and background gray levels; S4. Construct a detail enhancement lookup table, perform Gamma non-linear stretching on the normalized grayscale values, and construct a brightness weight lookup table to determine the detail weights based on the normalized brightness values. S5. Based on the grayscale allocation lookup table, detail enhancement lookup table, and brightness weight lookup table, perform lookup table fusion for each pixel to obtain the fusion result; S6. Linearly normalize the fusion result and map it to a finite bit-width output grayscale range to obtain the output image.

[0013] Preferably, the specific content of step S2 is as follows: When the histogram count of the gray level is not less than the preset support threshold, the frequency support of the gray level is determined to be valid. The number of gray levels whose frequency in the neighborhood of the histogram supports the existence of a gray level is counted. When the number of gray levels is not less than the preset minimum clustering number, the spatial clustering of gray levels is determined to be valid. When the histogram count of the gray level is not less than the intensity threshold, the relative intensity of the gray level is determined to be valid. The intensity threshold is determined by the support threshold and the intensity protection factor. When the frequency support of the gray level is valid, and the spatial aggregation or relative intensity is valid, the gray level is determined to be a valid gray level. Set the counts of invalid gray levels to zero, and retain the original counts of valid gray levels to obtain the valid histogram.

[0014] Preferably, step S2 further includes reducing the support threshold to 1 when the total number of gray levels that support the frequency of the entire image is less than 2, that is, using the histogram count of gray levels being greater than zero as the criterion for frequency support.

[0015] Preferably, in step S3, the method for constructing the grayscale allocation lookup table is as follows: Take the logarithm of each count value in the effective histogram and perform a log-weighted sum to obtain a log-weighted histogram; Calculate the cumulative distribution function of the log-weighted histogram and normalize it, then construct a gray-level allocation lookup table.

[0016] Preferably, in step S4, the enhanced detail lookup table is as follows:

[0017] in, For normalized grayscale values, For Gamma correction parameters, Enhance the lookup table for more details; The brightness weight lookup table is as follows:

[0018] in, For normalized brightness values, Based on the weighting coefficient, The brightness attenuation coefficient is... Use a lookup table for brightness weights.

[0019] Preferably, in step S5, for each pixel, the grayscale allocation result is obtained by querying the grayscale allocation lookup table with the logarithmic grayscale value, and the detail enhancement lookup table and the brightness weight lookup table are queried with the normalized grayscale value to obtain the detail enhancement result and the detail weight. The grayscale allocation result is superimposed with the detail enhancement result weighted by the detail weight to obtain the fusion result.

[0020] Preferably, in step S5, the fusion result obtained is as follows:

[0021] in, For the result of logarithmic field transformation, For normalized images, A lookup table is used to assign grayscale values. To enhance the lookup table for more details, For brightness weight lookup table, For the fusion result; In step S6, the output image obtained is:

[0022] in, and These represent the minimum and maximum values ​​of the fused result at the actual gray levels in the image, respectively. To output the image.

[0023] A high dynamic range infrared image adaptive processing system for complex scenes, based on the aforementioned high dynamic range infrared image adaptive processing method for complex scenes, includes: an image input module, a logarithmic domain transformation module, a grayscale histogram processing module, an effective histogram processing module, a grayscale allocation lookup table construction module, a detail enhancement lookup table module, a brightness weight lookup table module, a fusion module, and a mapping output module. Image input module, used to input high-resolution deep thermal infrared images; The logarithmic domain transformation module is used to perform a pixel-by-pixel logarithmic domain transformation on the input high-bit-depth thermal infrared image to obtain a logarithmic domain image. The grayscale histogram processing module is used to normalize the logarithmic domain image and then calculate the histogram of each grayscale level. The effective histogram processing module is used to filter effective gray levels based on the frequency support, spatial clustering and relative intensity of each gray level in the gray-level histogram to obtain an effective histogram, thereby reducing the occupancy of the output gray-level space by interfering gray levels. The grayscale allocation lookup table construction module is used to logarithmically weight the effective histogram and construct the grayscale allocation lookup table based on the cumulative distribution function of the logarithmically weighted histogram to adjust the output expression ratio of target-related grayscale and background grayscale. The detail enhancement lookup table module is used to perform Gamma non-linear stretching on normalized gray values ​​based on a pre-built detail enhancement lookup table. The luminance weight lookup table module is used to determine the detail weights based on the normalized luminance values ​​according to a pre-built luminance weight lookup table. The fusion module is used to perform table lookup fusion for each pixel based on the grayscale allocation lookup table, detail enhancement lookup table, and brightness weight lookup table to obtain the fusion result. The mapping output module is used to linearly normalize the fusion result and map it to a finite bit-width output grayscale range to obtain the output image.

[0024] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned method for adaptive processing of high dynamic infrared images in complex scenes.

[0025] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement a high dynamic range infrared image adaptive processing method for complex scenes.

[0026] As can be seen from the above technical solutions, compared with the prior art, the present invention discloses a high dynamic range infrared image adaptive processing method and system for complex scenes, which has the following beneficial effects: (1) This invention performs effective gray level screening for effective content expression of images. It determines the effectiveness of gray levels based on frequency support, spatial clustering and relative intensity, reduces the occupation of the 8-bit output gray space by noise, bad pixels, local abnormal brightness and fire interference gray levels, and retains real strong targets and strong edges through intensity protection criteria. (2) The present invention focuses on the target-background relationship expression. After logarithmically weighting the effective histogram, the cumulative distribution mapping is constructed. This reduces the occupation of the output gray space by the large area of ​​background gray and preserves the expression space of the target-related gray clusters, which is conducive to improving the distinguishability between the target and the background. (3) The present invention focuses on the expression of the target's own details by using a detail enhancement lookup table and a brightness weight lookup table to maintain the target's edges, textures and dark and weak levels, so that dark areas and weak target areas can obtain strong detail contributions, while constraining the enhancement of background details, bright hot spots and firelight areas, and reducing the risk of over-enhancement in 8-bit output. (4) The present invention uses table lookup mapping as the main operation. Effective gray level filtering is completed only in the gray level dimension with linear complexity. The lookup table has a small capacity and can be stored on the chip. It is convenient to implement in real time on various edge computing hardware in a data block parallel manner. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0028] Figure 1 This is a schematic diagram of the overall process of a high dynamic range infrared image adaptive processing method for complex scenes provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the original histogram provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of histogram processing for the effective content representation of an image provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of histogram processing for expressing the relationship between image target and background provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the Gamma nonlinear stretching curve of the detail enhancement lookup table in the target self-detail representation provided in the embodiments of the present invention; Figure 6 This is a schematic diagram of the brightness weight lookup table in the target's own detail representation provided in this embodiment of the invention. Figure 7 This is a schematic diagram of lookup table collaboration provided in an embodiment of the present invention. Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] Example 1 This invention discloses a high dynamic range infrared image adaptive processing method for complex scenes, such as... Figure 1 As shown, it specifically includes: S1. Perform a logarithmic domain transformation on each pixel of the input high-bit deep thermal infrared image to obtain a logarithmic domain image, and then normalize it before calculating the histogram of each gray level. S2. Based on the frequency support, spatial clustering and relative intensity of each gray level in the gray-level histogram, the effective gray levels are selected to obtain the effective histogram, thereby reducing the occupancy of the output gray-level space by interfering gray levels; S3. Log-weight the effective histogram and construct a gray-level allocation lookup table based on the cumulative distribution function of the log-weighted histogram to adjust the output ratio of target-related gray levels and background gray levels; S4. Construct a detail enhancement lookup table, perform Gamma non-linear stretching on the normalized grayscale values, and construct a brightness weight lookup table to determine the detail weights based on the normalized brightness values. S5. Based on the grayscale allocation lookup table, detail enhancement lookup table, and brightness weight lookup table, perform lookup table fusion for each pixel to obtain the fusion result; S6. Linearly normalize the fusion result and map it to a finite bit-width output grayscale range to obtain the output image.

[0031] In this embodiment, a logarithmic domain transformation is performed pixel-by-pixel on the input image:

[0032] in, To input thermal infrared images in coordinates Pixel value at that location, The result of the logarithmic field transformation; Logarithmic transformation is used to compress the span of high grayscale regions while preserving the relative differences in low grayscale regions; subsequently, the logarithmic domain image is normalized and its grayscale levels are statistically analyzed. Histogram .

[0033] To further implement the above technical solution, the specific content of step S2 is as follows: When the histogram count of the gray level is not less than the preset support threshold, the frequency support of the gray level is determined to be valid. grayscale If the frequency support is valid, the support mask is set to 1; otherwise, it is set to 0.

[0034] in, For support mask, grayscale Histogram count, The support threshold ranges from 2 to 4. The number of gray levels whose frequency in the neighborhood of the histogram supports the existence of a gray level is counted. When the number of gray levels is not less than the preset minimum clustering number, the spatial clustering of gray levels is determined to be valid. Specifically, to describe the spatial clustering relationship between gray levels, for normalized gray levels... In the Centered on, with radius Within the gray-level neighborhood, count the number of gray levels that satisfy the frequency support condition, and define the statistical result as spatial clustering. ;

[0035]

[0036] in, The radius of aggregation ranges from 1 to 3. The minimum cluster number is used to determine the minimum number of gray levels that satisfy the frequency support condition within the gray level neighborhood. Spatial clustering mask; spatial clustering degree By calculating the difference of the prefix sums of the support mask, and pre-calculating the prefix sum array of the support mask, the number of gray levels in the neighborhood of any gray level that support the frequency can be obtained in constant time by the difference of the prefix sums, so that the overall computational complexity of effective gray level screening is linear to the number of gray levels. When the histogram count of the gray level is not less than the intensity threshold, the relative intensity of the gray level is determined to be valid. The intensity threshold is determined by the support threshold and the intensity protection factor.

[0037] in, The intensity threshold, The strength protection factor ranges from 4 to 16. For relative intensity masking; When the frequency support of the gray level is valid, and the spatial aggregation or relative intensity is valid, the gray level is determined to be a valid gray level.

[0038] in, and They represent logical AND and logical OR, respectively. For validity mask; Set the count of invalid gray levels to zero, and retain the original count of valid gray levels to obtain the valid histogram;

[0039] in, For a valid histogram.

[0040] To further implement the above technical solution, step S2 also includes reducing the support threshold to 1 when the total number of gray levels that support the frequency of the whole image is less than 2, that is, using the histogram count of gray levels being greater than zero as the criterion for frequency support to be established.

[0041] At the level of effective image content representation, this embodiment improves the representation quality of real thermal infrared scene content through effective gray level screening. This level makes a comprehensive judgment based on the frequency of occurrence, neighborhood continuity and intensity of gray levels, retains stable and continuous scene gray levels, gray levels corresponding to real strong targets and strong edges, and reduces the occupation of noise, bad pixels, isolated abnormal brightness and fire light interference gray levels on the overall visual hierarchy and 8-bit output gray level space.

[0042] To further implement the above technical solution, the method for constructing the grayscale allocation lookup table in step S3 is as follows: For effective histograms Each count value is logarithmically divided and logarithmically weighted to obtain a logarithmically weighted histogram;

[0043] in, For logarithmically weighted histograms; Calculate the cumulative distribution function of the log-weighted histogram and normalize it, then construct a gray-level allocation lookup table;

[0044] in, Let be the cumulative distribution function of the log-weighted histogram. and These are its minimum and maximum values, respectively. Use a lookup table to assign grayscale values.

[0045] At the level of target-background relationship representation, this embodiment adjusts the proportion of target-related grayscale and background grayscale in the 8-bit output grayscale space by target-background grayscale allocation. This level is based on logarithmic weighting and cumulative distribution mapping of effective histograms to reduce the dominance of large-area background peaks on the output grayscale space, retain the mapping range of target-related grayscale clusters, and improve the grayscale distinguishability between the target and the background.

[0046] To further implement the above technical solution, in step S4, the enhanced detail lookup table is as follows:

[0047] in, The normalized grayscale value has a range of values. , This is the Gamma correction parameter, with a value ranging from 0.2 to 0.5. Enhance the lookup table for more details; The brightness weight lookup table is as follows:

[0048] in, For normalized brightness values, The basic weighting coefficient has a value range of 0.4 to 0.6. This is the brightness attenuation coefficient, with a value ranging from 0.2 to 0.3. A lookup table for brightness weights; When a pixel is located in a bright area, the detail weight is small, which is used to suppress the over-enhancement of background details, hot spots in bright areas, firelight, or saturated areas; when a pixel is located in a dark area or a weak target area, the detail weight is large, which is used to improve the discernibility of target edges, textures, and dark and weak levels.

[0049] To further implement the above technical solution, in step S5, for each pixel, the grayscale allocation result is obtained by querying the grayscale allocation lookup table with the logarithmic grayscale value, and the detail enhancement lookup table and the brightness weight lookup table are obtained by querying the normalized grayscale value respectively. The grayscale allocation result and the detail enhancement result weighted by the detail weight are superimposed to obtain the fusion result.

[0050] To further implement the above technical solution, the fusion result obtained in step S5 is as follows:

[0051]

[0052] in, For the result of logarithmic field transformation, For normalized images, A lookup table is used to assign grayscale values. To enhance the lookup table for more details, For brightness weight lookup table, For grayscale allocation results, For detail weighting, To enhance the results for details, For the fusion result; In step S6, the output image obtained is:

[0053] in, and These represent the minimum and maximum values ​​of the fused result at the actual gray levels in the image, respectively. To output the image.

[0054] At the level of target detail representation, this embodiment further preserves target edges, textures, and dark levels by maintaining target details. Based on the determination of effective image content representation and target-background relationship representation, this level expands the dark area level through Gamma nonlinear stretching and constrains the enhancement amplitude of background details, bright hot spots, and firelight areas by weights that decrease with brightness. After the separation between the target and the background is improved, the recognizability of the target's internal structure in the 8-bit output is maintained.

[0055] In practical applications, the input thermal infrared image in step S1 can be processed as a whole, in blocks, or in other data organization methods. To speed up data processing, the input thermal infrared image can be divided into multiple image blocks, and local logarithmic domain transformation and local histogram statistics can be performed separately for each block. The local histograms are then merged into a global histogram. Based on the global histogram, effective gray level filtering, logarithmic weighting, and gray level allocation lookup table construction are performed, and a global gray level allocation lookup table is constructed. Based on the gray level allocation lookup table, detail enhancement lookup table, and brightness weight lookup table, the respective image blocks are processed in parallel. After the lookup table fusion is completed, the block results are output and then stitched together according to the original image positions to form a complete output image.

[0056] Example 2 This embodiment uses a thermal infrared image with an input resolution of 640×512 and a bit depth of 14 bits as an example to illustrate a high dynamic range infrared image adaptive processing method for a complex scene according to Embodiment 1.

[0057] The grayscale range of the input thermal infrared image is 0~16383, and the grayscale range of the output image is 0~255. To meet the effective expression requirements when converting 14-bit information to 8-bit output, the relationship between effective content and interference content, the relationship between target and background are processed in sequence, while preserving the target edge and internal texture.

[0058] The input image is transformed pixel by pixel using a logarithmic domain transformation. This process compresses the span of high grayscale regions while preserving the relative differences in low grayscale regions. Subsequently, the logarithmic domain image is normalized and its grayscale histogram is calculated.

[0059] like Figure 2 As shown, the original histogram typically contains three types of gray levels: a high-count main peak corresponding to a large area of ​​background, continuous gray-level clusters related to the target, and isolated gray levels formed by noise, bad pixels, local abnormal brightness, or firelight interference.

[0060] This embodiment selects effective gray levels based on three criteria: frequency support, spatial clustering, and relative intensity, ensuring that the output image prioritizes representing realistic, continuous, and stable thermal infrared scene content; a support threshold is then used. Aggregation radius Minimum cluster number Strength protection factor Frequency support is valid when the gray level count is not less than 2; spatial clustering is valid when the number of gray levels in its neighborhood that support frequency is not less than 2; relative intensity is valid when the gray level count is not less than 16; after comprehensive judgment, the validity mask is obtained. and effective histogram ,like Figure 3 As shown, the count of invalid gray levels is set to zero.

[0061] like Figure 4 As shown, in this embodiment, the logarithm of the effective histogram is taken, the high count corresponding to the background main peak is compressed, and the mapping interval of the target-related gray-level cluster is preserved. Then, its cumulative distribution function is calculated and normalized to construct a gray-level allocation lookup table to determine the position of each gray level in the 8-bit output space, so that the space occupied by the background main peak is reduced and the target-related gray-level cluster remains continuously distinguishable, thereby adjusting the gray-level expression ratio of the target and the background.

[0062] Based on the established representation of effective image content and target-background relationship, a detail enhancement lookup table is constructed, such as... Figure 5 As shown, for normalized gray values Gamma stretching is performed in this embodiment. A value of 0.3 is used to expand the dark area layers and improve the recognizability of the target edges and textures, so as to preserve the internal structure of the target.

[0063] Construct a brightness weight lookup table, such as Figure 6 As shown in this embodiment Take 0.5, The value is set to 0.25. The detail weight is determined based on the normalized brightness value. The dark area has a larger weight, which contributes more to the target detail. The bright area has a smaller weight, which is used to suppress the over-enhancement of background details, hot spots in bright areas and firelight areas.

[0064] For each pixel, three lookup tables are queried and superimposed to obtain the fused result, such as... Figure 7 As shown, the grayscale allocation lookup table determines the output grayscale range of the target and background based on the effective histogram and logarithmic weighting results; the detail enhancement lookup table provides detail increments for target edges, textures, and dark levels; the brightness weight lookup table constrains these detail increments according to normalized brightness, reducing excessive enhancement of background details, bright hot spots, and firelight areas.

[0065] The fusion result is linearly normalized, and then linearly normalized again according to the minimum and maximum values ​​of the gray levels that actually appear in the image, and mapped to the output gray range to obtain an 8-bit output image.

[0066] Example 3 A high dynamic infrared image adaptive processing system for complex scenes, based on a high dynamic infrared image adaptive processing method for complex scenes, includes: an image input module, a logarithmic domain transformation module, a gray-level histogram processing module, an effective histogram processing module, a gray-level allocation lookup table construction module, a detail enhancement lookup table module, a brightness weight lookup table module, a fusion module, and a mapping output module. Image input module, used to input high-resolution deep thermal infrared images; The logarithmic domain transformation module is used to perform a pixel-by-pixel logarithmic domain transformation on the input high-bit-depth thermal infrared image to obtain a logarithmic domain image. The grayscale histogram processing module is used to normalize the logarithmic domain image and then calculate the histogram of each grayscale level. The effective histogram processing module is used to filter effective gray levels based on the frequency support, spatial clustering and relative intensity of each gray level in the gray-level histogram to obtain an effective histogram, thereby reducing the occupancy of the output gray-level space by interfering gray levels. The grayscale allocation lookup table construction module is used to logarithmically weight the effective histogram and construct the grayscale allocation lookup table based on the cumulative distribution function of the logarithmically weighted histogram to adjust the output expression ratio of target-related grayscale and background grayscale. The detail enhancement lookup table module is used to perform Gamma non-linear stretching on normalized gray values ​​based on a pre-built detail enhancement lookup table. The luminance weight lookup table module is used to determine the detail weights based on the normalized luminance values ​​according to a pre-built luminance weight lookup table. The fusion module is used to perform table lookup fusion for each pixel based on the grayscale allocation lookup table, detail enhancement lookup table, and brightness weight lookup table to obtain the fusion result. The mapping output module is used to linearly normalize the fusion result and map it to a finite bit-width output grayscale range to obtain the output image.

[0067] Example 4 A computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements a method for adaptive processing of high dynamic range infrared images in complex scenes.

[0068] Example 5 A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements a high dynamic range infrared image adaptive processing method for complex scenes.

[0069] The processor is edge computing hardware, which can be a multi-core digital signal processor, a field-programmable gate array, an embedded neural network processor, a graphics processor, or a heterogeneous computing chip containing any two or more of the above processing units.

[0070] In practical applications, a high dynamic range infrared image adaptive processing method for complex scenes can be implemented by using whole-frame processing, block processing or other data organization methods, depending on the processor type, memory storage structure, parallelism and data path. The block processing is used to adapt to the storage access and processing unit organization of the DSP. Specifically, the input thermal infrared image is divided into multiple image blocks, which are then distributed to multiple parallel processing units for processing. Each parallel processing unit performs local logarithmic domain transformation and local histogram statistics. The local histograms are merged into a global histogram, and effective gray level filtering, logarithmic weighting, and gray level allocation lookup table construction are performed based on the global histogram. The constructed global gray level allocation lookup table is distributed to each parallel processing unit. Each parallel processing unit processes its own image block in parallel based on the gray level allocation lookup table, detail enhancement lookup table, and brightness weight lookup table. After completing the lookup table fusion, the block results are output and then stitched together according to the original image positions to form a complete output image.

[0071] Furthermore, the detail enhancement lookup table and the brightness weight lookup table are independent of the input image content, are pre-calculated before processing, and reside in the on-chip high-speed memory of the edge computing hardware; the grayscale allocation lookup table is constructed online based on the global log-weighted histogram of the current frame; since the three lookup tables have small capacities, they can reside as a whole in the on-chip high-speed memory, thereby reducing the external memory access bandwidth requirements.

[0072] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0073] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for adaptive processing of high dynamic range infrared images in complex scenes, characterized in that, include: S1. Perform a logarithmic domain transformation on each pixel of the input high-bit deep thermal infrared image to obtain a logarithmic domain image, and then normalize it before calculating the histogram of each gray level. S2. Based on the frequency support, spatial clustering and relative intensity of each gray level in the gray-level histogram, the effective gray levels are selected to obtain the effective histogram, thereby reducing the occupancy of the output gray-level space by interfering gray levels; S3. Log-weight the effective histogram and construct a gray-level allocation lookup table based on the cumulative distribution function of the log-weighted histogram to adjust the output ratio of target-related gray levels and background gray levels; S4. Construct a detail enhancement lookup table, perform Gamma non-linear stretching on the normalized grayscale values, and construct a brightness weight lookup table to determine the detail weights based on the normalized brightness values. S5. Based on the grayscale allocation lookup table, detail enhancement lookup table, and brightness weight lookup table, perform lookup table fusion for each pixel to obtain the fusion result; S6. Linearly normalize the fusion result and map it to a finite bit-width output grayscale range to obtain the output image; The specific content of step S2 is as follows: When the histogram count of the gray level is not less than the preset support threshold, the frequency support of the gray level is determined to be valid. grayscale If the frequency support is valid, the support mask is set to 1; otherwise, it is set to 0. in, For support mask, grayscale Histogram count, Support threshold; The number of gray levels whose frequency in the neighborhood of the histogram supports the existence of a gray level is counted. When the number of gray levels is not less than the preset minimum clustering number, the spatial clustering of gray levels is determined to be valid. Spatial clustering for: in, Where is the aggregation radius, The minimum cluster number is used to determine the minimum number of gray levels that satisfy the frequency support condition within the gray level neighborhood. Spatial aggregation mask; When the histogram count of the gray level is not less than the intensity threshold, the relative intensity of the gray level is determined to be valid. The intensity threshold is determined by the support threshold and the intensity protection factor. in, The intensity threshold, As a strength protection factor, For relative intensity masking; When the frequency support of the gray level is valid, and the spatial aggregation or relative intensity is valid, the gray level is determined to be a valid gray level. in, and They represent logical AND and logical OR, respectively. For validity mask; Set the count of invalid gray levels to zero, and retain the original count of valid gray levels to obtain the valid histogram; in, For a valid histogram; In step S4, the detail enhancement lookup table is as follows: in, For normalized grayscale values, For Gamma correction parameters, Enhance the lookup table for more details; The brightness weight lookup table is as follows: in, For normalized brightness values, Based on the weighting coefficient, The brightness attenuation coefficient is... A lookup table for brightness weights; In step S5, for each pixel, the grayscale allocation lookup table is queried with the logarithmic grayscale value to obtain the grayscale allocation result, and the detail enhancement lookup table and the brightness weight lookup table are queried with the normalized grayscale value to obtain the detail enhancement result and the detail weight. The grayscale allocation result and the detail enhancement result weighted by the detail weight are superimposed to obtain the fusion result. The fusion result obtained is: in, For the result of logarithmic field transformation, For normalized images, A lookup table is used to assign grayscale values. To enhance the lookup table for more details, For brightness weight lookup table, For the fusion result; In step S6, the output image obtained is: in, and These represent the minimum and maximum values ​​of the fused result at the actual gray levels in the image, respectively. To output the image.

2. The adaptive processing method for high dynamic range infrared images in complex scenes as described in claim 1, characterized in that, Step S2 also includes reducing the support threshold to 1 when the total number of gray levels that support the frequency of the entire image is less than 2, that is, using the histogram count of gray levels being greater than zero as the criterion for frequency support.

3. The adaptive processing method for high dynamic range infrared images in complex scenes as described in claim 1, characterized in that, In step S3, the method for constructing the grayscale allocation lookup table is as follows: Take the logarithm of each count value in the effective histogram and perform a log-weighted sum to obtain a log-weighted histogram; Calculate the cumulative distribution function of the log-weighted histogram and normalize it, then construct a gray-level allocation lookup table.

4. A high dynamic range infrared image adaptive processing system for complex scenes, characterized in that, A high dynamic range infrared image adaptive processing method for complex scenes according to any one of claims 1-3 includes: an image input module, a logarithmic domain transformation module, a grayscale histogram processing module, an effective histogram processing module, a grayscale allocation lookup table construction module, a detail enhancement lookup table module, a brightness weight lookup table module, a fusion module, and a mapping output module. Image input module, used to input high-resolution deep thermal infrared images; The logarithmic domain transformation module is used to perform a pixel-by-pixel logarithmic domain transformation on the input high-bit-depth thermal infrared image to obtain a logarithmic domain image. The grayscale histogram processing module is used to normalize the logarithmic domain image and then calculate the histogram of each grayscale level. The effective histogram processing module is used to filter effective gray levels based on the frequency support, spatial clustering and relative intensity of each gray level in the gray-level histogram to obtain an effective histogram, thereby reducing the occupancy of the output gray-level space by interfering gray levels. The grayscale allocation lookup table construction module is used to logarithmically weight the effective histogram and construct the grayscale allocation lookup table based on the cumulative distribution function of the logarithmically weighted histogram to adjust the output expression ratio of target-related grayscale and background grayscale. The detail enhancement lookup table module is used to perform Gamma non-linear stretching on normalized gray values ​​based on a pre-built detail enhancement lookup table. The luminance weight lookup table module is used to determine the detail weights based on the normalized luminance values ​​according to a pre-built luminance weight lookup table. The fusion module is used to perform table lookup fusion for each pixel based on the grayscale allocation lookup table, detail enhancement lookup table, and brightness weight lookup table to obtain the fusion result. The mapping output module is used to linearly normalize the fusion result and map it to a finite bit-width output grayscale range to obtain the output image.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements a high dynamic range infrared image adaptive processing method for complex scenes as described in any one of claims 1 to 3.

6. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements a high dynamic range infrared image adaptive processing method for complex scenes as described in any one of claims 1 to 3.

Images