Method for enhancing and displaying low-light remote sensing images

Abstract

This invention discloses a low-light remote sensing image enhancement and display method to address the problem that existing technologies amplify noise while increasing contrast and redistributing brightness in low-light images. The method includes: parsing the original bitstream data to separate the original image from auxiliary data; extracting the integration time and image tube gain from the auxiliary data, and finding the relative radiometric calibration file from a preset calibration file; extracting the corresponding correction coefficients pixel-by-pixel from the relative radiometric calibration file, and performing pixel-by-pixel correction, median noise reduction filtering, and normalization on the original image to obtain a double-precision floating-point image, followed by multi-scale adaptive gamma correction; converting the corrected image into a double-precision integer image, and sequentially performing logarithmic stretching, normalization, and contrast-limited adaptive histogram equalization to obtain a new image; finally, converting the new image into a grayscale image adapted to the display to achieve image enhancement and display.

Description

Technical Field

[0001] This invention relates to image enhancement and display processing methods, specifically to an enhanced display method for scene-by-scene cataloging and browsing maps of low-light remote sensing images based on multi-scale adaptive gamma correction, and an enhanced display method for thumbnails of standard products of low-light remote sensing images. Background Technology

[0002] High-resolution Earth observation, using satellites as observation platforms, primarily utilizes optical imaging systems to acquire large-scale, high-precision, and multi-layered information about the Earth's surface. It is a modern strategic high-tech means for addressing a series of major issues, including environmental monitoring, and has become a field of intense development and competition among countries worldwide. However, existing high-resolution visible light remote sensing imaging systems are limited by the response sensitivity caused by the photosensitive mechanism of visible light detectors. High signal-to-noise ratio images can generally only be obtained when ground illumination is high, between 9 AM and 4 PM. Effective methods for high-resolution, high signal-to-noise ratio imaging under low-light or extremely low-light conditions, such as dawn / dusk, moonlit nights, or even starlight, still lack effective means.

[0003] Generally, low-light conditions refer to illumination conditions where the ground illuminance is less than 0.01 lux. When performing high-resolution visible light remote sensing imaging under typical low-light conditions such as a full moon, quarter moon, or even a crescent moon, the higher the spatial resolution, the fewer photons reach the detector pixels. Therefore, current CCD or CMOS detectors operating in orbit both domestically and internationally exhibit severe sensitivity deficiencies during nighttime imaging, with most only capable of imaging light-emitting areas and struggling to capture details in dark areas outside of light-emitting regions. ICCD or ICMOS detectors, which couple an image intensifier to a CCD or CMOS sensor, can significantly improve the detection sensitivity of weak signals in dark areas through the image tube's multiplication function. Furthermore, they do not amplify the dark current noise and readout noise inherent in the CCD or CMOS detector itself. They also feature automatic overload noise suppression, automatic brightness control, and automatic protection against strong light. Therefore, they can balance high sensitivity with a large dynamic range, making them an effective means of acquiring high-resolution low-light remote sensing images.

[0004] The working principle of ICCD or ICMOS detectors is as follows: Incident photons are first converted into initial photoelectrons by the photocathode. These initial photoelectrons are then amplified by a factor of millions through the secondary electron emission effect of the microchannel plate. Finally, the significantly enhanced electron cloud bombards the fluorescent screen under high voltage, achieving low-light enhancement. At this point, the CCD or CMOS detector, cascaded with the image intensifier, receives the light field amplified by the image intensifier, thus significantly improving its low-light detection capability. For example... Figure 1As shown, a comparison of the signal-to-noise ratio (SNR) of several candidate low-light detectors is presented. The following conclusions can be drawn: ① No detector can surpass the upper limit of the SNR given by the ideal camera. Among them, only ICMOS and ICCD detectors have performance closest to that of the ideal detector under dim lighting conditions; ② ICMOS and ICCD detectors perform best under low-light conditions; ③ Only ICMOS and ICCD detectors show relatively balanced performance under both strong and low light conditions. Therefore, using ICCD or ICMOS detectors to improve the image quality of low-light remote sensing has engineering value.

[0005] However, even when using ICCD or ICMOS detectors for low-light imaging, while the high gain of the image tube can significantly amplify sparse photons and thus acquire information from dark areas, the acquired low-light images still exhibit characteristics such as low contrast, low brightness, and low signal-to-noise ratio. Furthermore, due to the significant difference in dynamic range between illuminated and unilluminated areas, directly displaying the original image either reveals almost no useful information or only information from the illuminated areas. Histogram analysis shows that the histogram of a typical low-light image is mainly concentrated in low grayscale regions. Not only is the grayscale distribution of the entire image particularly concentrated, but concentrated distributions also form in areas with extremely small and extremely large grayscale values. This not only makes target scene identification and analysis difficult but also easily leads to misjudgments about the presence of useful remote sensing information in low-light images. Therefore, employing appropriate enhancement display methods to reduce noise in low-light images, suppress overexposure in bright areas, enhance details in dark areas, and improve both brightness and contrast in low-light images is crucial for maximizing the performance of low-light remote sensing image data.

[0006] Currently, traditional enhancement methods for low-light remote sensing images can be broadly categorized into two main types: spatial domain and transform domain. Spatial domain image enhancement methods are more common. These methods directly manipulate the grayscale of pixels in the image, primarily including histogram equalization (HE), adaptive histogram equalization (AHE), contrast-limited adaptive histogram equalization (CLAHE), and gamma correction. Because the HE algorithm applies the same transform to the entire image, it cannot adapt to contrast variations in different regions, especially when the image includes very dark areas. This can lead to an excessive increase in brightness in the processed image without effectively improving the overall grayscale dynamic range. Furthermore, partial merging of effective pixels can result in image information loss. AHE and CLAHE algorithms are improvements on the ordinary histogram equalization algorithm. They change the image contrast by calculating the histograms of multiple local regions and redistributing brightness. However, AHE can sometimes excessively increase local contrast, leading to image distortion. CLAHE performs better in this regard and is currently considered a relatively better contrast enhancement algorithm. Gamma correction expands the brightness of low-grayscale dark areas by compressing high-grayscale pixels. However, ordinary gamma correction, similar to the HE algorithm, uses the same gamma transform coefficient for each pixel, thus failing to adapt to the brightness differences in different areas of low-light images. Furthermore, whether it's the HE algorithm, AHE algorithm, or CLAHE algorithm combined with gamma correction, while increasing contrast and redistributing brightness in low-light images, they all amplify noise. Therefore, improving contrast in low-light images often requires simultaneous noise reduction filtering.

[0007] In summary, any existing single image enhancement and display method has its own limitations and is highly dependent on the application scenario. Therefore, there is an urgent need to develop an image enhancement and display method that can simultaneously achieve exposure suppression in bright areas, detail enhancement in dark areas, and signal-to-noise ratio improvement. Summary of the Invention

[0008] The purpose of this invention is to solve the technical problem that existing image enhancement and display methods amplify noise while increasing the contrast and redistributing brightness of low-light images, and to provide a low-light remote sensing image enhancement and display method.

[0009] To achieve the above objectives, the technical solution provided by this invention is as follows:

[0010] A method for enhancing the display of low-light remote sensing images, used for enhancing the display of scene-by-scene cataloging and browsing maps of low-light remote sensing images, is characterized by including the following steps:

[0011] Step 1: Parse the raw bitstream data of the low-light remote sensing image according to the preset remote sensing image format protocol to separate the raw image and auxiliary data;

[0012] Step 2: Extract the integration time and image tube gain from the auxiliary data respectively, and use the corresponding integration time-image tube gain combination as an index to find the relative radiometric calibration file corresponding to the integration time-image tube gain combination from the preset calibration file.

[0013] Step 3: Extract the corresponding correction coefficients pixel by pixel from the relative radiometric calibration file, and perform pixel-by-pixel correction on the original image based on the correction coefficients;

[0014] Step 4: Perform median noise reduction filtering and normalization on the original image after pixel-by-pixel correction to obtain a double-precision floating-point image.

[0015] Step 5: Perform multi-scale adaptive gamma correction on the double-precision floating-point image to obtain the corrected image;

[0016] Step 6: Convert the corrected image into a double-precision integer image, and then perform logarithmic stretching, normalization, and contrast-limited adaptive histogram equalization on it in sequence to obtain a new image.

[0017] Step 7: Convert the new image into a grayscale image that can be adapted to the display, thereby completing the enhanced display of the low-light remote sensing image scene classification and browsing map.

[0018] Furthermore, in step 1, the low-light remote sensing image is a single-channel grayscale image with a quantization bit depth of not less than 8 bits.

[0019] Furthermore, step 5 specifically includes:

[0020] Step 5.1: Convert the double-precision floating-point image to the original image. and the original image Perform the inversion operation to obtain the inverted image. ;

[0021] Step 5.2: Use the OTU algorithm to obtain the original images respectively. With inverted image The segmentation threshold is determined, and the average threshold of the two thresholds is calculated.

[0022] Step 5.3: Set up a deltas array, a winsizes array, a storage array, and a weight array, each containing M elements, where M ≥ 3; the sum of each element in the weight array is 1, and the weights are arranged in descending order.

[0023] Step 5.4, based on the deltas array, winsizes array, storage array, and weight array described in step 5.3, combines the average threshold to apply the average threshold to the inverted state image. Adaptive gamma correction is performed to obtain adaptive gamma-corrected images at different scales, and then weighted averages are applied to obtain the corrected image.

[0024] Furthermore, step 5.4 specifically includes:

[0025] Step 5.4.1, define the initial iteration count t and the maximum iteration count M, 1≤t≤M;

[0026] Step 5.4.2, let the array array Combined with the average threshold obtained in step 5.2, the maximum gamma correction coefficient and the minimum gamma correction coefficient corresponding to each pixel in the array currentdelta are calculated.

[0027] Step 5.4.3: Construct a two-dimensional Gaussian filter kernel with the array currentwinsize as the window size, and apply it to the inverted state image. The filtered image is obtained by performing Gaussian filtering.

[0028] Step 5.4.4: Based on the maximum and minimum gamma correction coefficients obtained in step 5.4.2, and the filtered image obtained in step 5.4.3, the inverted state image is calculated. The gamma correction coefficient for each pixel in the image;

[0029] Step 5.4.5: Based on the gamma correction coefficients obtained in step 5.4.4, process the inverted state image. Perform adaptive gamma correction processing to obtain pixel-adaptive gamma-corrected images and store them in a storage array;

[0030] Step 5.4.6: Normalize the pixel-adaptive gamma-corrected image and invert it again to obtain the adaptive gamma-corrected image corresponding to the current iteration. ;

[0031] Step 5.4.7, let the array And determine the relationship between t and M;

[0032] like Then let Then return to step 5.4.2;

[0033] like Then the array Each element in [t] is weighted and averaged according to the corresponding element in the weight array to obtain the corrected image.

[0034] Furthermore, in step 6, the double-precision integer image The expression is:

[0035] ;

[0036] In the formula, The image is the corrected image from step 5.4.7; DOUBLE represents the double-precision forced conversion operator; N represents the original quantization bit depth of the image, and N≥10;

[0037] Image after logarithmic stretching The expression is:

[0038] ;

[0039] Among them, log D This represents the logarithmic operator with base D, where D > 1; c is a constant and c ≥ 1;

[0040] In step 7, the grayscale image The expression is:

[0041] ;

[0042] Where UINT represents the integer type casting operator; S represents the maximum number of bits for display quantization, and ; This represents the new image obtained after contrast-limited adaptive histogram equalization.

[0043] Furthermore, this invention also provides another method for enhancing the display of low-light remote sensing images, used for enhancing the display of thumbnails of standard low-light remote sensing image products, characterized by including the following steps:

[0044] Step 1: Parse the raw bitstream data of the low-light remote sensing image according to the preset remote sensing image format protocol to separate the raw image and auxiliary data;

[0045] Step 2: Extract the integration time and image tube gain from the auxiliary data respectively, and use the corresponding integration time-image tube gain combination as an index to find the relative radiometric calibration file corresponding to the integration time-image tube gain combination from the preset calibration file.

[0046] Step 3: Extract the corresponding correction coefficients pixel by pixel from the relative radiometric calibration file, and perform pixel-by-pixel correction on the original image based on the correction coefficients;

[0047] Step 4: Starting with the original image after pixel-by-pixel correction, standard product production is carried out to obtain the standard product image;

[0048] Step 5: Perform median noise reduction filtering and normalization on the standard product image in sequence to obtain a double-precision floating-point image;

[0049] Step 6: Perform multi-scale adaptive gamma correction on the double-precision floating-point image to obtain the corrected image;

[0050] Step 7: Convert the corrected image into a double-precision integer image, and then perform logarithmic stretching, normalization, and contrast-limited adaptive histogram equalization on it in sequence to obtain a new image.

[0051] Step 8: Convert the new image into a grayscale image that can be adapted to the display, thereby completing the enhanced display of the thumbnail of the low-light remote sensing image standard product.

[0052] Furthermore, in step 1, the low-light remote sensing image is a single-channel grayscale image with a quantization bit depth of not less than 8 bits.

[0053] Furthermore, step 6 specifically includes:

[0054] Step 6.1: Convert the double-precision floating-point image to the original image. and the original image Perform the inversion operation to obtain the inverted image. ;

[0055] Step 6.2: Use the OTU algorithm to obtain the original images respectively. With inverted image The segmentation threshold is determined, and the average threshold of the two thresholds is calculated.

[0056] Step 6.3: Set up a deltas array, a winsizes array, a storage array, and a weight array, each containing M elements, where M ≥ 3; the sum of each element in the weight array is 1, and the weights are arranged in descending order.

[0057] Step 6.4, based on the deltas array, winsizes array, storage array, and weight array in step 6.3, combines the average threshold to apply the average threshold to the inverted state image. Adaptive gamma correction is performed to obtain adaptive gamma-corrected images at different scales, and then weighted averages are applied to obtain the corrected image.

[0058] Furthermore, step 6.4 specifically includes:

[0059] Step 6.4.1, define the initial iteration count t and the maximum iteration count M, 1≤t≤M;

[0060] Step 6.4.2, let the array array Combined with the average threshold obtained in step 6.2, the maximum gamma correction coefficient and the minimum gamma correction coefficient corresponding to each pixel in the array currentdelta are calculated.

[0061] Step 6.4.3: Construct a two-dimensional Gaussian filter kernel with the array currentwinsize as the window size, and apply it to the inverted state image. The filtered image is obtained by performing Gaussian filtering.

[0062] Step 6.4.4: Based on the maximum and minimum gamma correction coefficients obtained in step 6.4.2, and the filtered image obtained in step 6.4.3, the inverted state image is calculated. The gamma correction coefficient for each pixel in the image;

[0063] Step 6.4.5: Based on the gamma correction coefficients obtained in step 6.4.4, process the inverted state image. Perform adaptive gamma correction processing to obtain pixel-adaptive gamma-corrected images and store them in a storage array;

[0064] Step 6.4.6: Normalize the pixel-adaptive gamma-corrected image and invert it again to obtain the adaptive gamma-corrected image corresponding to the current iteration. ;

[0065] Step 6.4.7, let the array And determine the relationship between t and M;

[0066] like Then let Then return to step 6.4.2;

[0067] like Then the array Each element in [t] is weighted and averaged according to the corresponding element in the weight array to obtain the corrected image.

[0068] Furthermore, in step 7, the double-precision integer image The expression is:

[0069] ;

[0070] In the formula, The corrected image obtained in step 6.4.7; DOUBLE represents the double-precision forced conversion operator; N represents the original quantization bit depth of the image, and N≥10;

[0071] Image after logarithmic stretching The expression is:

[0072] ;

[0073] Among them, log D This represents the logarithmic operator with base D, where D > 1; c is a constant and c ≥ 1;

[0074] In step 8, the grayscale image The expression is:

[0075] ;

[0076] Where UINT represents the integer type casting operator; S represents the maximum number of bits for display quantization, and ; This represents the new image obtained after contrast-limited adaptive histogram equalization.

[0077] The beneficial effects of this invention are as follows:

[0078] 1. The low-light remote sensing image enhancement and display method provided by the present invention can not only achieve the purpose of suppressing exposure in bright areas, improving details in dark areas and improving signal-to-noise ratio, but also overcome the high dependence of a single enhancement algorithm on the scene, so that the low-light remote sensing image enhancement and display method can be used in different lighting and target scenes such as night, dawn and dusk, city, and suburbs.

[0079] 2. The low-light remote sensing image enhancement and display method provided by the present invention uses relative radiometric correction as the front-end processing flow for low-light remote sensing image enhancement and display. While eliminating the striping effect caused by the large difference in gain response curves when using ICCD or ICMOS at high gain, it can also perform preliminary adjustment of the contrast and brightness of the low-light remote sensing image.

[0080] 3. The low-light remote sensing image enhancement and display method provided by the present invention couples three methods: multi-scale adaptive gamma correction, small-base logarithmic stretching, and contrast-limited adaptive histogram equalization, effectively overcoming the limitations of single-method application scenarios.

[0081] 4. The low-light remote sensing image enhancement and display method provided by this invention has very low spatial and temporal complexity, does not rely on third-party databases, can be used across platforms, and has negligible latency in processing 4K high-definition images. Therefore, it can be used not only for PC processing but also has the potential for embedded applications. Attached Figure Description

[0082] Figure 1 This is a graph showing the relationship between the signal-to-noise ratio (SNR) of a typical existing low-light detector and the number of incident photons. The horizontal axis represents the number of incident photons received by each pixel, and the vertical axis represents the SNR.

[0083] Figure 2 This is a flowchart illustrating Embodiment 1 of the low-light remote sensing image enhancement and display method of the present invention;

[0084] Figure 3This is a flowchart illustrating Embodiment 2 of the low-light remote sensing image enhancement and display method of the present invention;

[0085] Figure 4 This is a flowchart of the multi-scale adaptive gamma correction method in step 5 of embodiment 1 of the low-light remote sensing image enhancement and display method of the present invention. Detailed Implementation

[0086] In recent years, with the explosive growth of computing power in computer hardware, deep learning-based low-light image enhancement has ushered in a period of rapid development. In particular, improving the quality of low-light images has become a highly challenging research direction at top international image processing conferences. However, deep learning-based low-light image enhancement is highly dependent on low-light image datasets, but current training of deep learning networks for low-light image enhancement severely lacks support from typical aerospace remote sensing low-light image datasets. Most existing limited low-light image datasets use ordinary CCD or CMOS detectors as imaging devices, and their noise patterns and response characteristics differ significantly from those of ICCD and ICMOS detectors, as well as from the imaging modes of spaceborne low-light remote sensing. Therefore, although traditional low-light image enhancement algorithms require accurate imaging models and suffer from insufficient robustness, their advantage of not relying on image datasets makes it valuable to propose enhancement and display methods suitable for low-light remote sensing images based on in-depth exploration and optimized combination of traditional low-light image enhancement algorithms.

[0087] Based on this, the present invention delves into the potential of traditional low-light image enhancement methods and, combined with the characteristics of low-light remote sensing imaging, proposes a low-light remote sensing image enhancement and display method based on multi-scale adaptive gamma correction. This method is performed in the spatial domain, and the main calculations are all point operations. It has the characteristics of being simple and fast. It can be used on PCs, such as ground application systems paired with satellites, or embedded in FPGAs or DSPs for real-time processing.

[0088] To make the objectives, advantages, and features of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Those skilled in the art should understand that these embodiments are merely used to explain the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0089] Example 1

[0090] like Figure 2 As shown, this embodiment provides a method for enhancing the display of low-light remote sensing images, used to enhance the brightness and contrast of the scene-by-scene cataloging view of low-light remote sensing images, including the following steps:

[0091] Step 1: Parse the raw bitstream data of the low-light remote sensing image according to the preset remote sensing image format protocol to separate the raw image. The image contains auxiliary data, which is used to extract imaging-related parameters, while the original image contains source data that needs to be enhanced for display.

[0092] The low-light remote sensing images described in this embodiment are acquired by ICCD or ICMOS detectors. The low-light remote sensing images are single-channel grayscale images with a quantization bit depth of not less than 8 bits.

[0093] Step 2: Extract the integration time and image tube gain from the auxiliary data respectively, and use the corresponding integration time-image tube gain combination as an index to find the relative radiometric calibration file corresponding to the integration time-image tube gain combination from the preset calibration file, thereby completing the preparation work before the enhanced display processing.

[0094] Step 3: Extract the corresponding correction coefficients a and b pixel by pixel from the relative radiometric calibration file, and apply these correction coefficients to the original image. Perform pixel-by-pixel correction to obtain the corrected original image. ,Right now: .

[0095] This embodiment utilizes a relative radiometric calibration file to perform relative radiometric correction on low-light remote sensing images, thereby eliminating inconsistencies in the responses of adjacent pixels within each detector and between different detectors. This improves the relative radiometric quality of low-light images while also allowing for preliminary adjustments to contrast and brightness.

[0096] Step 4: Correct the original image pixel by pixel. The image is obtained by performing median noise reduction filtering. Then, normalization is performed to obtain a double-precision floating-point image. ,in .

[0097] Because the low-light image before enhancement has significant noise, directly performing enhancement without denoising filtering would further amplify the noise, thus degrading the enhanced display effect. Therefore, after completing the preliminary work, denoising filtering is performed first to improve the signal-to-noise ratio. If a certain time delay is acceptable, non-local mean filtering (NLM) and bilateral regularization filtering algorithms can better preserve edge details while reducing noise. If low latency and high real-time performance are required, simple small window midpoint denoising filtering can also improve the signal-to-noise ratio. Using the denoised image as input, the conditions for executing the enhancement display algorithm are met.

[0098] Step 5, for double-precision floating-point images Multi-scale adaptive gamma correction is performed to obtain the corrected image. Gamma correction is a classic contrast enhancement algorithm that achieves contrast enhancement by compressing high-grayscale pixels and expanding the brightness of low-grayscale dark areas through exponentiation. However, it typically uses the same gamma correction factor for the entire image, making it difficult to simultaneously achieve both bright area compression and dark area expansion. In reality, the illuminance in bright areas of low-light remote sensing images, such as areas with artificial light, can reach hundreds or even thousands of lux, while the illuminance in unlit areas may be far less than 0.01 lux. This significant difference in brightness naturally requires gamma correction to be adaptively set for the brightness level of each pixel's neighborhood. Therefore, this embodiment introduces multi-scale local brightness measurement based on inversion states and proposes multi-scale adaptive gamma correction to adapt to the vast dynamic range of low-light remote sensing images.

[0099] Combination Figure 4 As shown, the process of multi-scale adaptive gamma correction is as follows:

[0100] Step 5.1, make the double-precision floating-point image Original image and the original image Perform the inversion operation to obtain the inverted image. ,Right now Since targets of interest in low-light images are often hidden in dark, low-contrast areas, this step involves inverting the low-light remote sensing image, swapping the background and foreground, and treating the dark targets as foreground to highlight the information in the dark areas of the image.

[0101] Step 5.2: Use the OTU algorithm (Maximum Inter-Class Variance Method) to obtain the original images respectively. Segmentation threshold level and inverted state image The segmentation threshold invlevel is calculated, and the average of the two is used to obtain the average threshold avglevel, which is the expected value of brightness enhancement.

[0102] Step 5.3: Set up a deltas array, winsizes array, storage array and weight array with M elements respectively, where M≥3.

[0103] in, ;

[0104] ;

[0105] Storage array This is used to store the adaptive gamma-corrected image at each scale;

[0106] weight array The sum of all elements in this array is 1, and the weights are arranged in descending order.

[0107] Step 5.4, based on the deltas array, winsizes array, storage array, and weight array described in step 5.3, combines the average threshold avglevel to process the inverted state image. Adaptive gamma correction is performed to obtain adaptive gamma-corrected images at different scales, and then weighted averages are applied to obtain the corrected image.

[0108] Step 5.4.1: Define the initial iteration count t (initial value is 1) and the maximum iteration count M, 1≤ t≤M, and enter the following loop.

[0109] Step 5.4.2, let the array array And, combined with the average threshold avglevel obtained in step 5.2, calculate the maximum gamma correction coefficient gammamax and the minimum gamma correction coefficient gammamin for each pixel in the currentdelta array, that is:

[0110] ;

[0111] .

[0112] Step 5.4.3: Construct a two-dimensional Gaussian filter kernel with the array currentwinsize as the window size, and apply it to the inverted state image. Gaussian filtering is performed to obtain the filtered image. Each pixel value represents a value derived from a given image. The size is used as a measure of the local brightness level.

[0113] This step utilizes windows of different sizes, such as 3×3, 5×5, and 7×7, to calculate the brightness level of the local neighborhood of each pixel in the inverted image using Gaussian filtering. This serves as the basis for generating adaptive gamma coefficients that match the brightness level of the pixel neighborhood. In the original image space, the larger the gamma coefficient, the more significantly dark areas are brightened, and the more significantly overexposed bright areas are compressed. However, in the inverted image, the opposite is true: the larger the gamma coefficient, the darker the dark areas and the brighter the bright areas.

[0114] Step 5.4.4: Based on the maximum gamma correction coefficient (gammamax) and minimum gamma correction coefficient (gammamin) obtained in step 5.4.2, and the filtered image obtained in step 5.4.3... The inverted state image is calculated. The gamma correction coefficient gammamat for each pixel is:

[0115] .

[0116] It is obvious that the I of the local pixel neighborhood in the inverted image localmean The higher the brightness value, the higher the corresponding gamma coefficient. Since the dark areas in the inverted image correspond to the bright areas in the original image, further enhancing the bright areas in the inverted image is actually enhancing the dark areas in the original image. Furthermore, in both the inverted and original images, the difference between the maximum and minimum gamma correction coefficients determines the contrast; the greater the difference, the stronger the contrast.

[0117] Step 5.4.5: Based on the gamma correction coefficients (gammamat) obtained in step 5.4.4, perform the following steps on the inverted state image: Perform adaptive gamma correction processing, that is, invert the image. Each pixel in the array is obtained according to step 5.4.4. The corresponding elements are exponentially operated on to obtain the pixel-adaptive gamma-corrected image. And store it in a storage array; pixel-adaptive gamma-corrected image The expression is:

[0118] ;

[0119] Where (i, j) represents the two-dimensional index of the pixel, and POW represents the exponentiation operator.

[0120] Step 5.4.6, perform pixel-adaptive gamma correction on the image. Normalization is performed, and then the image is inverted again to obtain the adaptive gamma-corrected image corresponding to the current iteration. ,Right now .

[0121] Step 5.4.7, let the array And determine the relationship between t and M;

[0122] like Then let Then return to step 5.4.2;

[0123] like Then the array Each element in [t] is determined by a weight array. The corresponding elements are weighted and averaged to obtain the output image. ,Right now:

[0124] ;

[0125] The output image That is, the corrected image. .

[0126] After performing multi-scale adaptive gamma correction, the overall brightness and contrast of the low-light image were further improved based on the relative radiometric correction. However, due to the large dynamic range of the low-light image, further compression of bright areas is still necessary. Research shows that the zero-frequency intensity of the image's Fourier spectrum is much higher than other high-frequency intensities. Logarithmic stretching can suppress the zero-frequency intensity and enhance the high-frequency intensity, thereby revealing the high-frequency details of the image spectrum.

[0127] Step 6, convert the corrected image Convert to double-precision integer image ,in:

[0128] ;

[0129] In the formula, DOUBLE represents the double-precision forced conversion operator; N represents the original quantization bit depth of the image, and N≥10 (in this embodiment, N=16).

[0130] Next, the double-precision integer image Perform small-base logarithmic stretching to obtain the stretched image. ,Right now:

[0131] ;

[0132] In the formula, log D This represents a logarithmic operator with base D, where D > 1, and c is a small constant greater than or equal to 1 to avoid logarithmic aberrations. This operation is similar to the enhancement of low-light remote sensing images, which needs to consider both bright and dark areas. This embodiment introduces a method used for image spectral enhancement to the enhancement of low-light remote sensing images, which can further enhance the dark areas of the image after multi-scale adaptive gamma correction. The smaller the base of the logarithm used, the more significant the dark area enhancement effect.

[0133] Next, stretch the image Normalization is performed to obtain the image. .

[0134] Finally, regarding the image A new image is obtained by performing contrast-limited adaptive histogram equalization (CLAHE). Because logarithmic stretching severely compresses the bright areas of an image, the small difference in grayscale between pixels leads to a significant loss of contrast. Therefore, by using a contrast-limited adaptive histogram equalization method to perform block-based enhancement to improve contrast, it is possible to enhance the contrast of images that have undergone multi-scale adaptive gamma correction and small-base logarithmic stretching, which have significantly improved brightness.

[0135] Step 7, transfer the new image Convert to grayscale image that fits the monitor. This enables enhanced display of the low-light remote sensing image scene-by-scene cataloging and browsing map.

[0136] In this embodiment, Where UINT represents the integer type casting operator, and S represents the maximum number of bits displayed on the screen. For remote sensing images, this is usually... .

[0137] The core of the image enhancement display method in this embodiment is to couple three methods: multi-scale adaptive gamma correction, small-base logarithmic stretching, and contrast-limited adaptive histogram equalization, thereby overcoming the limitations of single-method application scenarios.

[0138] First, since the targets of interest in low-light images are mostly hidden in dark, low-contrast areas, the original low-light image is first inverted to swap the foreground and background and highlight the dark areas. Then, enhancement processing is performed on the inverted image.

[0139] Secondly, in the multi-scale adaptive gamma correction algorithm, firstly, based on the inverted state image, local brightness measurement is performed using Gaussian filtering based on different scale windows. Based on this, the gamma correction coefficient corresponding to the scale window and matching the brightness of the neighborhood of each pixel is calculated. Then, the gray value of each pixel is adaptively changed through gamma correction, thereby increasing the image brightness level and improving contrast. At the same time, the adaptive gamma correction results corresponding to different scale windows need to be weighted and averaged to obtain the low-light image after multi-scale adaptive gamma correction.

[0140] Furthermore, after the Fourier transform of an image, the intensity difference between zero-frequency and high-frequency intensities in the spectrum is extremely large. Without any processing, only the zero-frequency information in the spectrum is visible. Logarithmic stretching is generally used to display the high-frequency information of the image spectrum. The huge dynamic range between bright and dark areas in low-light remote sensing images is similar. Therefore, logarithmic stretching is particularly suitable for suppressing overexposure in bright areas and significantly increasing the brightness of dark areas in low-light remote sensing images. The smaller the logarithm base, the better the effect of increasing the brightness of dark areas.

[0141] Finally, while logarithmic stretching significantly improves the brightness of dark areas, the concentrated pixel grayscale distribution results in low contrast. Therefore, contrast-limited adaptive histogram equalization (HQE) is needed to enhance contrast. HQE offers advantages in block-based enhancement and suppressing excessive local contrast increases. Applying HQE again to an image whose brightness has been greatly improved after adaptive gamma correction and logarithmic stretching significantly enhances both brightness and contrast, thereby highlighting details in dark areas and achieving high-quality enhanced display of low-light images.

[0142] The image enhancement and display method of this embodiment not only achieves the goals of suppressing exposure in bright areas, enhancing details in dark areas, and improving the signal-to-noise ratio, but also overcomes the high dependence of single enhancement algorithms on the scene. This allows the low-light remote sensing image enhancement and display method to be applied to different lighting and target scenes, such as nighttime, dawn / dusk, urban areas, and suburbs. This enhancement and display method uses only the statistical characteristics of the low-light remote sensing image itself and the accompanying relative radiometric calibration file, without relying on any third-party algorithm libraries. Furthermore, this enhancement and display method runs on a PC and can process 4K high-definition images in real time with a latency of less than 20ms.

[0143] Example 2

[0144] like Figure 3 As shown, this embodiment provides a method for enhancing the display of low-light remote sensing images, used to enhance the brightness and contrast of thumbnails of standard low-light remote sensing image products, including the following steps:

[0145] Step 1: Parse the raw bitstream data of the low-light remote sensing image according to the preset remote sensing image format protocol to separate the raw image. With auxiliary data.

[0146] Step 2: Extract the integration time and image tube gain from the auxiliary data respectively, and use the corresponding integration time-image tube gain combination as an index to find the relative radiometric calibration file corresponding to the integration time-image tube gain combination from the preset calibration file.

[0147] Step 3: Extract the corresponding correction coefficients a and b pixel by pixel from the relative radiometric calibration file, and apply these correction coefficients to the original image. Perform pixel-by-pixel correction to obtain the corrected original image. ,Right now: .

[0148] Step 4, using the original image after pixel-by-pixel correction Starting with standard product production (i.e., undergoing processes such as sensor calibration, fusion, geometric calibration, and geometric fine calibration), a standard product image is obtained. .

[0149] Step 5, standard product image Median noise reduction filtering is performed sequentially to obtain the image. ;

[0150] Next, the image after median noise reduction filtering... Normalization is performed to obtain a double-precision floating-point image. , where 0≤ ≤1.

[0151] Step 6, for double-precision floating-point images Multi-scale adaptive gamma correction is performed to obtain the corrected image. The specific correction process is the same as step 5 of Example 1, and will not be repeated here.

[0152] Step 7, the corrected image Convert to double-precision integer image ,in, N represents the original quantization bit depth of the image, with a value of 16, and DOUBLE represents the double-precision cast operator.

[0153] Double precision integer image Perform small-base logarithmic stretching to obtain the stretched and enhanced image. ,in logD represents the logarithmic operator with base D, where D>1, and c is a small constant greater than or equal to 1 to avoid logarithmic aberrations.

[0154] Next, the image Normalization is performed to obtain the image. .

[0155] Finally, regarding the image A new image is obtained by performing contrast-limited adaptive histogram equalization. .

[0156] Step 8, transfer the new image Convert to grayscale image that fits the monitor. ,in Where UINT represents the integer type casting operator, and S represents the maximum number of bits displayed on the screen. For remote sensing images, this is usually... This enables enhanced display of thumbnails for standard low-light remote sensing image products.

[0157] 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 the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein, and such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the present invention.

Claims

1. A method for enhancing the display of low-light remote sensing images, used for enhancing the display of scene-by-scene cataloging and browsing maps of low-light remote sensing images, characterized in that, Includes the following steps: Step 1: Parse the raw bitstream data of the low-light remote sensing image according to the preset remote sensing image format protocol to separate the raw image and auxiliary data; Step 2: Extract the integration time and image tube gain from the auxiliary data respectively, and use the corresponding integration time-image tube gain combination as an index to find the relative radiometric calibration file corresponding to the integration time-image tube gain combination from the preset calibration file. Step 3: Extract the corresponding correction coefficients pixel by pixel from the relative radiometric calibration file, and perform pixel-by-pixel correction on the original image based on the correction coefficients; Step 4: Perform median noise reduction filtering and normalization on the original image after pixel-by-pixel correction to obtain a double-precision floating-point image. Step 5: Perform multi-scale adaptive gamma correction on the double-precision floating-point image to obtain the corrected image; Step 6: Convert the corrected image into a double-precision integer image, and then perform logarithmic stretching, normalization, and contrast-limited adaptive histogram equalization on it in sequence to obtain a new image. Step 7: Convert the new image into a grayscale image that can be adapted to the display, thereby completing the enhanced display of the low-light remote sensing image scene classification and browsing map.

2. The low-light remote sensing image enhancement and display method according to claim 1, characterized in that: In step 1, the low-light remote sensing image is a single-channel grayscale image with a quantization bit depth of not less than 8 bits.

3. The low-light remote sensing image enhancement and display method according to claim 1, characterized in that, Step 5 specifically involves: Step 5.1: Convert the double-precision floating-point image to the original image. and the original image Perform the inversion operation to obtain the inverted image. ; Step 5.2: Use the OTU algorithm to obtain the original images respectively. With inverted image The segmentation threshold is determined, and the average threshold of the two thresholds is calculated. Step 5.3: Set up a deltas array, a winsizes array, a storage array, and a weight array, each containing M elements, where M ≥ 3; the sum of each element in the weight array is 1, and the weights are arranged in descending order. Step 5.4, based on the deltas array, winsizes array, storage array, and weight array described in step 5.3, combines the average threshold to apply the average threshold to the inverted state image. Adaptive gamma correction is performed to obtain adaptive gamma-corrected images at different scales, and then weighted averages are applied to obtain the corrected image.

4. The low-light remote sensing image enhancement and display method according to claim 3, characterized in that, Step 5.4 specifically involves: Step 5.4.1, define the initial iteration count t and the maximum iteration count M, 1≤t≤M; Step 5.4.2, let the array array Combined with the average threshold obtained in step 5.2, the maximum gamma correction coefficient and the minimum gamma correction coefficient corresponding to each pixel in the array currentdelta are calculated. Step 5.4.3: Construct a two-dimensional Gaussian filter kernel with the array currentwinsize as the window size, and apply it to the inverted state image. The filtered image is obtained by performing Gaussian filtering. Step 5.4.4: Based on the maximum and minimum gamma correction coefficients obtained in step 5.4.2, and the filtered image obtained in step 5.4.3, the inverted state image is calculated. The gamma correction coefficient for each pixel in the image; Step 5.4.5: Based on the gamma correction coefficients obtained in step 5.4.4, process the inverted state image. Perform adaptive gamma correction processing to obtain pixel-adaptive gamma-corrected images and store them in a storage array; Step 5.4.6: Normalize the pixel-adaptive gamma-corrected image and invert it again to obtain the adaptive gamma-corrected image corresponding to the current iteration. ; Step 5.4.7, let the array And determine the relationship between t and M; like Then let Then return to step 5.4.2; like Then the array Each element in [t] is weighted and averaged according to the corresponding element in the weight array to obtain the corrected image.

5. The low-light remote sensing image enhancement and display method according to claim 4, characterized in that: In step 6, double-precision integer image The expression is: ； In the formula, The image after correction in step 5.4.7; DOUBLE represents the double-precision type coercion operator; N represents the original quantization bits of the image, and N≥10; Image after logarithmic stretching The expression is: ； Among them, log D This represents the logarithmic operator with base D, where D > 1; c is a constant and c ≥ 1; In step 7, the grayscale image The expression is: ； Where UINT represents the integer type casting operator; S represents the maximum number of bits for display quantization, and ; This represents the new image obtained after contrast-limited adaptive histogram equalization.

6. A method for enhancing the display of low-light remote sensing images, used for enhancing the display of thumbnails of standard low-light remote sensing image products, characterized in that, Includes the following steps: Step 1: Parse the raw bitstream data of the low-light remote sensing image according to the preset remote sensing image format protocol to separate the raw image and auxiliary data; Step 2: Extract the integration time and image tube gain from the auxiliary data respectively, and use the corresponding integration time-image tube gain combination as an index to find the relative radiometric calibration file corresponding to the integration time-image tube gain combination from the preset calibration file. Step 3: Extract the corresponding correction coefficients pixel by pixel from the relative radiometric calibration file, and perform pixel-by-pixel correction on the original image based on the correction coefficients; Step 4: Starting with the original image after pixel-by-pixel correction, standard product production is carried out to obtain the standard product image; Step 5: Perform median noise reduction filtering and normalization on the standard product image in sequence to obtain a double-precision floating-point image; Step 6: Perform multi-scale adaptive gamma correction on the double-precision floating-point image to obtain the corrected image; Step 7: Convert the corrected image into a double-precision integer image, and then perform logarithmic stretching, normalization, and contrast-limited adaptive histogram equalization on it in sequence to obtain a new image. Step 8: Convert the new image into a grayscale image that can be adapted to the display, thereby completing the enhanced display of the thumbnail of the low-light remote sensing image standard product.

7. The low-light remote sensing image enhancement and display method according to claim 6, characterized in that: In step 1, the low-light remote sensing image is a single-channel grayscale image with a quantization bit depth of not less than 8 bits.

8. The low-light remote sensing image enhancement and display method according to claim 6, characterized in that, Step 6 specifically involves: Step 6.1: Convert the double-precision floating-point image to the original image. and the original image Perform the inversion operation to obtain the inverted image. ; Step 6.2: Use the OTU algorithm to obtain the original images respectively. With inverted image The segmentation threshold is determined, and the average threshold of the two thresholds is calculated. Step 6.3: Set up a deltas array, a winsizes array, a storage array, and a weight array, each containing M elements, where M ≥ 3; the sum of each element in the weight array is 1, and the weights are arranged in descending order. Step 6.4, based on the deltas array, winsizes array, storage array, and weight array in step 6.3, combines the average threshold to apply the average threshold to the inverted state image. Adaptive gamma correction is performed to obtain adaptive gamma-corrected images at different scales, and then weighted averages are applied to obtain the corrected image.

9. The low-light remote sensing image enhancement and display method according to claim 8, characterized in that, Step 6.4 specifically involves: Step 6.4.1, define the initial iteration count t and the maximum iteration count M, 1≤t≤M; Step 6.4.2, let the array array Combined with the average threshold obtained in step 6.2, the maximum gamma correction coefficient and the minimum gamma correction coefficient corresponding to each pixel in the array currentdelta are calculated. Step 6.4.3: Construct a two-dimensional Gaussian filter kernel with the array currentwinsize as the window size, and apply it to the inverted state image. The filtered image is obtained by performing Gaussian filtering. Step 6.4.4: Based on the maximum and minimum gamma correction coefficients obtained in step 6.4.2, and the filtered image obtained in step 6.4.3, the inverted state image is calculated. The gamma correction coefficient for each pixel in the image; Step 6.4.5: Based on the gamma correction coefficients obtained in step 6.4.4, process the inverted state image. Perform adaptive gamma correction processing to obtain pixel-adaptive gamma-corrected images and store them in a storage array; Step 6.4.6: Normalize the pixel-adaptive gamma-corrected image and invert it again to obtain the adaptive gamma-corrected image corresponding to the current iteration. ; Step 6.4.7, let the array And determine the relationship between t and M; like Then let Then return to step 6.4.2; like Then the array Each element in [t] is weighted and averaged according to the corresponding element in the weight array to obtain the corrected image.

10. The low-light remote sensing image enhancement and display method according to claim 9, characterized in that: In step 7, double-precision integer image The expression is: ； In the formula, The corrected image obtained in step 6.4.7; DOUBLE represents the double-precision type coercion operator; N represents the original quantization bits of the image, and N≥10; Image after logarithmic stretching The expression is: ； Among them, log D This represents the logarithmic operator with base D, where D > 1; c is a constant and c ≥ 1; In step 8, the grayscale image The expression is: ； Where UINT represents the integer type casting operator; S represents the maximum number of bits for display quantization, and ; This represents the new image obtained after contrast-limited adaptive histogram equalization.

Images