Television camera image enhancement method for low-light environments

Abstract

The present application belongs to the technical field of image enhancement, and particularly relates to a television camera image enhancement method for low-illumination environment, comprising: acquiring initial pixel values of each color channel at the current moment and the previous moment; extracting an initial illumination component and combining spatial structure fluctuation and time difference to construct a basic space-time fluctuation value; modulating the mean value of the initial illumination component in the neighborhood to obtain a space-time illumination fluctuation intensity; constructing an adaptive diffusion coefficient according to the space-time illumination fluctuation intensity and dynamically weighting the initial illumination component to obtain a target illumination component; constructing a dynamic compensation factor by using the relative intensity of the target illumination component and the global distribution extreme value, and then generating a final enhanced image. The present application can effectively suppress environmental noise and eliminate halo artifacts, and adaptively improve the contrast and visual fidelity of the texture in extremely dark areas.

Description

Technical Field

[0001] This invention relates to the field of image enhancement technology. More specifically, this invention relates to an image enhancement method for television cameras in low-light environments. Background Technology

[0002] When television cameras operate in low-light environments such as at night or in enclosed spaces, the captured video stream exhibits overall low brightness and environmental noise interference. Due to the scarcity of ambient photons, the signal in dark areas of the image is extremely weak, and the shot noise and dark current noise generated by the sensor exhibit high-frequency random fluctuations, causing real physical targets to be easily obscured by the background noise. To improve image quality, the industry typically employs image enhancement processing techniques to stretch the contrast and restore details in the video image.

[0003] Chinese patent document CN103593830B discloses a low-light video image enhancement method. This method maps a frame of a video image from the RGB color space to the HSI color space, extracts the original illuminance component from the original luminance component, calculates the original illuminance component using a multi-scale Retinex algorithm, and stretches the histogram to 0-255 to obtain the first illuminance component. An evaluation function is constructed to calculate the corresponding integral parameters, and the first illuminance component is processed using an integral formula to obtain the second illuminance component. Finally, the first luminance component is reconstructed by combining the reflection component and converted back to an RGB image. Furthermore, this method calculates the Euclidean distance between the histograms of two adjacent frames to determine whether to reuse the enhancement parameters of the previous frame for inter-frame enhancement of the video image. However, this patent document mainly relies on histogram stretching and global evaluation functions for brightness conversion, without deeply distinguishing between the extremely weak real physical texture in dark areas and sensor shot noise. While forcibly stretching the brightness of dark areas, it is easy to cause high-frequency noise to be amplified simultaneously. Moreover, this method only judges the macroscopic inter-frame similarity through the Euclidean distance of the histogram, which severs the difference between the motion of real physical targets and the simple thermal electron perturbation in local spatiotemporal space, resulting in the enhanced image still being easily contaminated by dynamic random noise.

[0004] In existing technologies, although some solutions attempt to enhance low-light images through multi-scale filtering or global histogram adjustment, the light distribution in low-light scenes is extremely uneven, and the high-frequency sensor noise is highly similar to weak physical textures in a single numerical value. Such isolated comparison logic, which relies on local brightness amplification or macroscopic inter-frame statistics, fails to dynamically identify the true physical source of local high-frequency abrupt signal changes. This results in the system being unable to accurately separate real dynamic targets from random environmental interference when processing complex light abrupt changes or extremely dark areas. When the brightness of dark areas is increased by stretching grayscale, the lack of a signal-to-noise separation mechanism inevitably leads to a proportional and sharp amplification of the noise at the same frequency. Furthermore, when using uniform-scale filtering and smoothing operations, the inability to adaptively block energy penetration across physical boundaries easily produces severe halo artifacts at the intersection of strong light and erases weak details, fundamentally restricting the visual reliability of television cameras in accurately capturing dynamic targets in low-light environments. Summary of the Invention

[0005] To address the technical problem that existing technologies cannot accurately separate the real target from random background noise, leading to a rapid amplification of noise, halo artifacts, and the erasure of subtle details during enhancement, this invention provides an image enhancement method for television cameras in low-light environments, comprising: S1: Obtain the initial pixel values ​​of each color channel in the current and previous time steps in the image sequence to be processed; S2: Based on the initial pixel values ​​of each color channel, extract the initial illumination components of the current and previous time steps, and construct the basic spatiotemporal fluctuation value of the current time step by combining spatial structure fluctuation and temporal difference; use the mean of the initial illumination components in the neighborhood as the basis of signal intensity to modulate the basic spatiotemporal fluctuation value to obtain the spatiotemporal illuminance fluctuation intensity of the current time step; S3: Construct an adaptive diffusion coefficient based on the global discrete characteristics of the spatiotemporal illuminance fluctuation intensity, and use the adaptive diffusion coefficient to dynamically weight the initial illumination components to obtain the target illumination component of the current time step; S4: Construct a dynamic compensation factor using the relative intensity of the target illumination component and the global distribution extreme value, and combine the initial pixel values ​​of the channels, the target illumination component, and the dynamic compensation factor to generate the final enhanced image.

[0006] This invention adds the temporal difference to the spatial gradient and controls it under a local illumination base, ensuring that high-frequency shot noise in dark areas is forcibly suppressed, resulting in a significant structural response from realistic bright moving targets and achieving physical separation of random background noise from real edge features. This invention utilizes an adaptive diffusion coefficient reflecting global discrete features to dynamically weight the initial illumination distribution, giving flat noise regions extremely high smoothness and cutting off energy leakage paths at the boundaries of strong light abrupt changes, thus blocking the generation of halo artifacts. This invention uses a dynamic compensation mechanism combining global distribution extrema with relative intensity to specifically enhance the extracted physical reflection characteristics, preventing overexposure in bright areas while achieving high-intensity stretching of subtle but realistic texture details in extremely dark areas. This overcomes the shortcomings of traditional fixed smoothing scales that easily amplify noise and cause artifacts, adaptively improving global contrast while effectively suppressing environmental noise, achieving a high degree of unity between realistic detail preservation and visual quality in low-light scenes.

[0007] Preferably, the step of extracting the initial illumination components at the current time and the previous time includes: obtaining the maximum value among the initial pixel values ​​of each color channel at the same spatial coordinate at the current time and the previous time, and using it as the initial illumination component at the corresponding spatial coordinate.

[0008] This invention constructs the initial illumination component by extracting the maximum value of the initial pixel values ​​of each color channel at the same spatial coordinates. This effectively captures and retains the most significant light source radiation information in the scene, avoiding illumination estimation deviations caused by the influence of differences in the reflectivity of the object's own color on a single channel. This provides a stable and physically meaningful illumination benchmark for the subsequent accurate extraction of spatiotemporal fluctuation features.

[0009] Preferably, the method for obtaining the spatial structure fluctuation is as follows: calculate the mean of the absolute values ​​of the differences between the initial illumination component at the current moment and the initial illumination component in a set spatial neighborhood, and use this as the spatial structure fluctuation value.

[0010] Preferably, constructing the basic spatiotemporal fluctuation value at the current moment includes: adding the absolute value of the time difference between the initial illumination component at the current moment and the previous moment to the spatial structure fluctuation value to obtain the basic spatiotemporal fluctuation value at the current moment.

[0011] This invention sums the absolute value of the inter-frame temporal difference, which reflects the true motion characteristics of the target, with the spatial structure fluctuation value, which reflects the physical edge structure. This allows the extracted basic spatiotemporal fluctuation value to simultaneously take into account both the dynamic changes and static abrupt changes in the video stream. This effectively compensates for the defect that single-frame spatial features are prone to misjudging random shot noise as physical structure, and achieves the initial spatiotemporal feature fusion of dynamic moving targets and local physical structures.

[0012] Preferably, the spatiotemporal illuminance fluctuation intensity satisfies the expression: ;in, Represents the spatial x-coordinate at the current time. with spatial ordinate The intensity of spatiotemporal illuminance fluctuations at a given location; Indicates the current time; Represents the spatial x-coordinate at the current time. with spatial ordinate The fundamental spatiotemporal fluctuation value at that location; Representing the current time in terms of spatial x-coordinate with spatial ordinate The mean of the initial illumination components within the 8-neighborhood of the center; This represents the dark field basis bias constant.

[0013] This invention utilizes the mean of the initial illumination component in the neighborhood and the dark field basis bias constant as the signal modulation basis to modulate the extracted basic spatiotemporal fluctuation value. In the pure noise region with an extremely dark background, the extremely small local mean will forcefully suppress the false fluctuation response caused by high-frequency shot noise. However, when encountering the edge of a real bright target, the significantly increased mean of the initial illumination component will rapidly amplify the fluctuation response, thereby separating the high-frequency dark field random noise from the real moving edge.

[0014] Preferably, the adaptive diffusion coefficient satisfies the expression: In the formula, Represents the spatial x-coordinate at the current time. with spatial ordinate The adaptive diffusion coefficient at the location; Indicates the current time; Represents the spatial x-coordinate at the current time. with spatial ordinate The intensity of spatiotemporal illuminance fluctuations at a given location; This represents the standard deviation of the spatiotemporal illuminance fluctuation intensity of all pixels in the image to be processed at the current moment. This represents an exponential function with the natural constant as its base.

[0015] This invention utilizes the spatiotemporal illumination fluctuation intensity and its global discrete standard deviation in the entire image to construct an adaptive diffusion coefficient with an exponential distribution. It can automatically generate continuous smooth weights based on the degree of spatiotemporal physical drastic changes in the region where the pixel is located, so that the adaptive diffusion coefficient can reflect the significance of local fluctuation features in the global illumination environment, providing a reliable dynamic control basis for subsequent adaptive optimization of target illumination components.

[0016] Preferably, the target illumination component satisfies the expression: ;in, Represents the spatial x-coordinate at the current time. with spatial ordinate The target illumination component at the location; Indicates the current time; Represents the spatial x-coordinate at the current time. with spatial ordinate The initial illumination component at the location; Represents the spatial x-coordinate at the current time. with spatial ordinate The adaptive diffusion coefficient at the location; Representing the current time in terms of spatial x-coordinate with spatial ordinate The mean of the initial illumination components within the 8-neighborhood of the center.

[0017] This invention utilizes an adaptive diffusion coefficient to dynamically adjust the fusion weight of the initial illumination component and its local mean. It imparts extremely high mean smoothing strength in flat dark fields or noisy regions with slight fluctuations, while directly preserving the initial illumination characteristics in high-brightness physical abrupt regions with drastic spatiotemporal fluctuations. This smoothing strategy, which adaptively switches based on physical characteristics, blocks energy leakage at the strong light interface and eliminates halo artifacts caused by a fixed smoothing scale.

[0018] Preferably, the dynamic compensation factor satisfies the expression: In the formula, Represents the spatial x-coordinate at the current time. with spatial ordinate Dynamic compensation factor at the location; Indicates the current time; Represents the spatial x-coordinate at the current time. with spatial ordinate The target illumination component at the location; This represents the maximum value of the target illumination component of all pixels in the image to be processed at the current moment; This represents the illuminance constraint parameter.

[0019] This invention uses the maximum value of the target illumination component of the entire frame image as the global normalization benchmark, and combines the relative illumination intensity of the pixel itself with the illumination constraint parameters to perform nonlinear power-law mapping. While adaptively raising the global dynamic compensation base, it gives the originally weak reflection signal in the dark area an extremely high stretching ratio, effectively breaking the dilemma that the tiny textures in the extremely dark area are completely submerged by the background noise, and improving the visual recognition of local targets.

[0020] Preferably, generating the final enhanced image includes: the enhanced pixel values ​​of each color channel satisfying the expression: ;in, Indicates the color channel at the current moment. Spatial x-coordinate with spatial ordinate Enhanced pixel values ​​at the location; Indicates the current time; Represents the spatial x-coordinate at the current time. with spatial ordinate Initial pixel value of the channel at that location; Indicates the color channel number; Represents the spatial x-coordinate at the current time. with spatial ordinate The target illumination component at the location; This represents the dark field basis bias constant; Represents the spatial x-coordinate at the current time. with spatial ordinate Dynamic compensation factor at the location; This represents the maximum value of the target illumination component of all pixels in the image to be processed at the current moment; the final enhanced image is generated based on the enhanced pixel values ​​of each color channel.

[0021] This invention utilizes a dynamic compensation factor to remodulate the stably stripped physical reflection signal, enabling the realistic and subtle textures in dark areas to obtain targeted gains. This achieves adaptive enhancement of contrast and suppression of background noise across the entire scene while preventing overexposure in highlight areas.

[0022] Preferably, the method for obtaining the illuminance constraint parameter includes: taking pictures in a standard darkroom environment using a standard multi-level grayscale test card to obtain the actual output brightness value of each grayscale test block, drawing a photoelectric conversion characteristic curve in combination with the standard reflectance corresponding to each grayscale test block, fitting the photoelectric conversion characteristic curve to obtain the power exponent, and setting the power exponent as the illuminance constraint parameter.

[0023] The beneficial effects of this invention are as follows: This invention combines inter-frame temporal difference with spatial neighborhood gradient in an additive manner, and uses the mean of the initial illumination component within the neighborhood as the physical modulation basis to construct the spatiotemporal illumination fluctuation intensity. This can forcefully suppress high-frequency shot noise in dark areas while preserving realistic bright dynamic targets, thus reversing the defect of traditional algorithms that misjudge high-frequency noise as physical texture. This invention constructs an adaptive diffusion coefficient based on the difference between spatiotemporal physical edges and flat noise regions, automatically adjusting the smoothing weight of the initial illumination component, breaking through the limitations of traditional fixed smoothing scales, and blocking the penetration of halo artifacts at high-contrast abrupt edges. This invention uses the global distribution extrema of the target illumination component to construct a dynamic compensation factor, adaptively amplifying the extracted physical reflection component. While effectively preventing overexposure in bright areas, it allows for high-intensity stretching of faint but realistic texture details in extremely dark areas. This invention achieves a high degree of unity between random noise suppression, artifact elimination, and global contrast enhancement in extremely low illumination environments, improving the visual reliability and image fidelity of television cameras capturing small dynamic targets in all weather conditions. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating the image enhancement method for television cameras in low-light environments according to the present invention. Figure 2 This is a schematic diagram of the image to be processed at the current moment; Figure 3 This is a schematic diagram illustrating the effect of the initial illumination component. Figure 4 A schematic diagram illustrating the effect of spatiotemporal illuminance fluctuation intensity; Figure 5 This is a schematic diagram illustrating the effect of adaptive diffusion coefficient. Figure 6 A schematic diagram illustrating the effect of the target illumination component; Figure 7 This is a schematic diagram illustrating the effect of the dynamic compensation factor. Figure 8 This is a schematic diagram illustrating the final enhanced image effect. Detailed Implementation

[0025] 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, not all, of the embodiments of the present invention. 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.

[0026] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0027] This invention discloses an image enhancement method for television cameras in low-light environments, referring to... Figure 1 This includes steps S1 to S4: S1. Obtain the initial pixel values ​​of each color channel in two consecutive frames.

[0028] Specifically, a television camera is used to acquire a sequence of images to be processed at multiple consecutive moments under extremely low illumination conditions. The television camera has a global exposure color detector, supports night imaging mode, and has an extremely low minimum illumination index.

[0029] The video frames of the image sequence to be processed are parsed into a two-dimensional matrix with effective pixels. The initial pixel values ​​of each color channel in the current time and the previous time are extracted from the image sequence to be processed, including the initial pixel values ​​of the red channel, the green channel, and the blue channel.

[0030] For example, Figure 2The image shows raw video frames captured in extremely low-light surveillance scenarios. The overall brightness of the image is extremely low, it contains severe shot noise, and it also hides faint reflective textures, making it difficult to discern target details with the naked eye.

[0031] S2. Calculate the initial illumination component and the spatiotemporal illuminance fluctuation intensity.

[0032] It should be noted that, due to the drastic random amplitude fluctuations caused by shot noise and dark current noise in image sensors under extremely low illumination conditions, relying solely on spatial neighborhood differences can easily lead to the misjudgment of random noise as real physical motion edges. Therefore, this invention weights and sums the temporal difference representing the target motion with the spatial neighborhood gradient representing the physical structure, and uses the mean of the initial illumination component in the current neighborhood as the physical modulation basis of the signal intensity to construct the spatiotemporal illumination fluctuation intensity. This ensures that high-frequency shot noise in the dark area is forcibly suppressed by the local low illumination basis, while the real bright moving target can produce a significant structural response, thus achieving the physical separation of noise and edge features.

[0033] Specifically, based on the initial pixel values ​​of the red channel, green channel, and blue channel in the image sequence to be processed at the current moment, the initial illumination components at the current moment are calculated: ; In the formula, Represents the spatial x-coordinate at the current time. with spatial ordinate The initial illumination component at the location; Indicates the current time; Represents the spatial x-coordinate at the current time. with spatial ordinate Initial pixel value of the red channel at that location; Represents the spatial x-coordinate at the current time. with spatial ordinate Initial pixel value of the green channel at that location; Represents the spatial x-coordinate at the current time. with spatial ordinate Initial pixel value of the blue channel at that location; This represents the maximum value function.

[0034] For example, Figure 3 This is a schematic diagram of the effect of the initial lighting component. The initial lighting component initially reveals the lighting outline in the scene, but it is still limited by the extremely low illumination environment, resulting in low overall contrast and obvious high-frequency shot noise.

[0035] Similarly, obtain the spatial x-coordinate of the previous time step. with spatial ordinate The initial illumination component at that location.

[0036] Furthermore, based on the initial illumination component at the current moment, the spatial structure fluctuation value at the current moment is calculated: ; In the formula, Represents the spatial x-coordinate at the current time. with spatial ordinate Spatial structure fluctuation value at the location; Represents the spatial x-coordinate at the current time. with spatial ordinate The initial illumination component at the location; Indicates the current time; Represents the x-coordinate of the neighborhood; Represents the ordinate of the neighborhood; Represents the x-coordinate of the neighborhood at the current time. and neighborhood ordinates The initial illumination component at the location; Indicates the absolute value symbol; Indicates the current pixel point The set of 8 neighboring pixel coordinates centered on the object. This invention utilizes spatial structure fluctuation values ​​to extract fluctuation features characterizing physical structures. When the difference in initial illumination components in the neighborhood increases, the spatial structure fluctuation value increases, indicating that the spatial coordinates at the current moment... and neighborhood coordinates There are fluctuations in physical structure.

[0037] Furthermore, based on the initial illumination components and spatial structure fluctuation values ​​at the current and previous moments, the fundamental spatiotemporal fluctuation value at the current moment is calculated: ; in, Represents the spatial x-coordinate at the current time. with spatial ordinate The fundamental spatiotemporal fluctuation value at that location; Represents the spatial x-coordinate at the current time. with spatial ordinate The initial illumination component at the location; Indicates the current time; Represents the spatial x-coordinate of the previous time step. with spatial ordinate The initial illumination component at the location; This represents the absolute value of the time difference between the initial illumination component at the current moment and the previous moment; Represents the spatial x-coordinate at the current time. with spatial ordinate The spatial structure fluctuation value at the location. When the absolute value of the time difference or the spatial structure fluctuation value increases, the basic spatiotemporal fluctuation value increases, thereby initially extracting the fluctuation characteristics containing the target motion and physical structure.

[0038] Furthermore, based on the fundamental spatiotemporal fluctuation values ​​and the initial illumination components, the spatiotemporal illuminance fluctuation intensity at the current moment is calculated: ; in, Represents the spatial x-coordinate at the current time. with spatial ordinate The intensity of spatiotemporal illuminance fluctuations at a given location; Indicates the current time; Represents the spatial x-coordinate at the current time. with spatial ordinate The fundamental spatiotemporal fluctuation value at that location; Representing the current time in terms of spatial x-coordinate with spatial ordinate The mean of the initial illumination components within the 8-neighborhood of the center; This represents the dark-field basis bias constant. Personnel can capture multiple frames of images in a completely dark environment and calculate the average value of the output pixel values ​​for each color channel as the dark-field basis bias constant to establish the lowest response base for physical thermal noise. In extremely dark monitoring scenarios, high-frequency shot noise manifests as isolated pixel flickering, which is not a real physical structure or moving target. Furthermore, the overall environment is under extremely low illumination, resulting in a very small mean of the initial illumination component in the neighborhood. The spatiotemporal illumination fluctuation intensity is significantly reduced due to the strong suppression by the physical basis. Conversely, when encountering a real moving target or a bright physical edge, the difference in the target's own illumination intensity and structure causes a significant increase in the basic spatiotemporal fluctuation value and the mean of the initial illumination component in the neighborhood, thus increasing the spatiotemporal illumination fluctuation intensity. This achieves the separation of bright physical edges from high-frequency noise in dark areas.

[0039] For example, Figure 4 This is a schematic diagram illustrating the effect of spatiotemporal illuminance fluctuation intensity. It can be seen that... Figure 4 It highlights the realistic physical edges and structural response of dynamically moving targets.

[0040] S3. Calculate the adaptive diffusion coefficient and extract the target illumination component.

[0041] It should be noted that in extremely low-light environments, the image signal output by the sensor is overwhelmed by a large amount of shot noise and thermal noise. Traditional fixed smoothing algorithms cannot distinguish whether local fluctuations originate from real physical edges or from random noise spikes. In order to achieve accurate denoising and edge protection, this invention reverts to the underlying statistical physical model of sensor noise, treats local spatiotemporal fluctuations as random variables, uses the Gaussian probability density distribution function to calculate the probability that the fluctuation belongs to pure random noise, and uses the opposite event of this probability as the basis for the allocation of adaptive smoothing weights, thereby achieving adaptive separation of signal and noise.

[0042] Specifically, in a flat background area of ​​a low-light monitoring scenario, the purely random noise caused by sensor thermal electron perturbations statistically follows a Gaussian normal distribution with zero mean. The spatial abscissa of the current moment... with spatial ordinate Spatiotemporal illuminance fluctuation intensity at the location Treating the observed fluctuation as a random variable, according to the central limit theorem and the standard Gaussian probability density function, the probability that this local fluctuation is entirely caused by pure random noise is... Satisfying the expression: ; In the formula, Represents the spatial x-coordinate at the current time. with spatial ordinate The probability value that the fluctuation at a certain point belongs to pure random noise; Represents an exponential function with the natural constant as its base; This indicates the intensity of the spatiotemporal illuminance fluctuation at the current moment; It represents the standard deviation of the spatiotemporal illumination fluctuation intensity of all pixels in the image to be processed at the current moment. This standard deviation statistically quantifies the global basic noise level of the entire image at the current moment.

[0043] Furthermore, after obtaining the probability that a local fluctuation belongs to noise, based on the axiom of opposite events in probability theory, this local fluctuation is not noise, but rather the signal confidence of a real, objectively existing physical edge or moving target, which is equal to 1 minus the probability that it belongs to noise. This invention directly defines this confidence level, which characterizes the probability of the existence of a real signal, as the adaptive diffusion coefficient. It satisfies the expression: ； In the formula, Represents the spatial x-coordinate at the current time. with spatial ordinate The adaptive diffusion coefficient at that location.

[0044] Furthermore, based on the initial illumination component and the adaptive diffusion coefficient, the target illumination component at the current moment is calculated: ; in, Represents the spatial x-coordinate at the current time. with spatial ordinate The target illumination component at the location; Indicates the current time; Represents the spatial x-coordinate at the current time. with spatial ordinate The initial illumination component at the location; Represents the spatial x-coordinate at the current time. with spatial ordinate The adaptive diffusion coefficient at the location; Representing the current time in terms of spatial x-coordinate with spatial ordinate The mean of the initial illumination components within the 8-neighborhood of the center.

[0045] The greater the spatiotemporal illuminance fluctuation intensity and the smaller the standard deviation of the spatiotemporal illuminance fluctuation intensity, the larger the adaptive diffusion coefficient, which makes the target illumination component closer to the initial illumination component. Conversely, when the adaptive diffusion coefficient approaches 0, the target illumination component is closer to the mean of the initial illumination component in the neighborhood. In nighttime monitoring scenarios with extremely low illumination, the spatiotemporal illuminance fluctuation intensity in flat dark areas and shot noise regions is extremely small, and the adaptive diffusion coefficient approaches 0. When calculating the target illumination component, the neighborhood mean dominates the generation of the target illumination component, giving the flat area extremely high smoothness, thus making the target illumination component closer to the mean of the initial illumination component in the neighborhood. However, when encountering real high-brightness illumination abrupt change edges, because the spatiotemporal illuminance fluctuation intensity is significantly greater than the global dispersion, the adaptive diffusion coefficient quickly approaches 1. When calculating the target illumination component, the original characteristics of the initial illumination component are directly retained, making the target illumination component closer to the initial illumination component. This invention dynamically weights the initial illumination component and its local mean by using an adaptive diffusion coefficient, which eliminates noise interference and cuts off the energy leakage path at the strong light interface, thus avoiding the generation of halo artifacts.

[0046] For example, Figure 5 This is a schematic diagram illustrating the effect of adaptive diffusion coefficient. Figure 5 In flat dark areas and low-illuminance noise regions, it exhibits a low adaptive diffusion coefficient response to guide the algorithm to use local mean for high-intensity smoothing, while at the edge of real high-brightness illumination abrupt change, it exhibits a high adaptive diffusion coefficient response to directly preserve the original characteristics of the initial illumination, thus realizing the adaptive spatial allocation of feature preservation and smoothing intensity. Figure 6This diagram illustrates the effect on the target illumination component, showcasing the illumination reconstruction result after dynamically weighting the initial illumination component and its local mean using an adaptive diffusion coefficient. Figure 6 It effectively eliminates high-frequency noise interference in dark areas, while maintaining the original illuminance distribution characteristics at the intersection of strong light and weak light, thus avoiding energy leakage and halo artifacts.

[0047] S4. Construct dynamic compensation factors and generate the final enhanced image.

[0048] It should be noted that in extremely low-light environments, the physical energy signals collected by television camera sensors are extremely weak. Even if the physical reflection signals reflecting object properties are extracted through the aforementioned steps, direct linear stretching and output will result in severe loss of detail and depth in dark areas due to the extremely low signal-to-noise ratio and highly collapsed energy distribution in the dark region. The core technical challenge lies in the fact that the photoelectric response of camera sensors is absolutely linear at the physical level, while the physiological perception of photon stimulation by human retinal neurons strictly follows a nonlinear power-law response. This mismatch between physical acquisition and physiological perception makes it impossible for conventional methods to restore the sense of depth that conforms to visual habits in dark environments. Therefore, this invention introduces a physiological response model of the human visual system for reverse reconstruction. By deriving a nonlinear power-law compensation relationship to generate a dynamic compensation factor, it accurately achieves nonlinear expansion of weak signals in extremely dark areas and adaptive clamping in bright areas at the physiological mapping constraint level.

[0049] Specifically, based on Stevens' power law in the theory of visual physiological perception and the grayscale response principle of standard display devices, the final visual brightness perceived by humans strictly follows a power-law distribution relationship with the input pixel physical electrical signal, and its fundamental physical response equation satisfies the expression: ; In the formula, This represents the final brightness perceived by the human eye; This represents the input pixel driving electrical signal; This represents the illuminance constraint parameter, characterizing the nonlinear properties of visual perception and the display medium. To ensure that weak physical features in low-illuminance environments are mapped without distortion to the linear range of human physiological perception after enhancement, targeted inverse pre-distortion compensation must be implemented in advance during signal processing. Let the current spatial coordinates be... The normalized relative physical illumination distribution at the location is In order to achieve visual linearization, it is mandatory to require Substitute the above-mentioned physical response fundamental equations and solve in reverse for the required pixel driving electrical signals. The necessary reverse mapping relationship is derived: .

[0050] Due to pixel drive electrical signals This invention characterizes the underlying energy reference that must be input into the display terminal to achieve ideal visual perception. To adaptively configure the nonlinearity of this energy reference during image processing, the present invention materializes the theoretical inverse mapping relationship as a dynamic compensation factor applied to illumination characteristics. In essence, the dynamic compensation factor is a spatially adaptive pre-distortion operator constructed to obtain the ideal pixel driving electrical signal. To establish a globally consistent brightness reference, the normalized relative physical illumination distribution is... Defined as local target illumination component Compared with the global maximum energy extremum of the current image Substituting the physical ratio into the inverse mapping relationship obtained above, the dynamic compensation factor at the current moment is derived. Satisfies the final expression: ; In the formula, Represents the spatial x-coordinate at the current time. with spatial ordinate Dynamic compensation factor at the location; Indicates the target illumination component; This represents the maximum value of the target illumination component for all pixels in the image to be processed at the current moment. It's important to note that when... When the current time is a completely black image, the dynamic compensation factor of all pixels in the current time image is set to 0. This represents the illuminance constraint parameter. It should be noted that when... At this time, the compensation function exhibits convex function characteristics, which in physical terms means that: when the target illumination component... In smaller, extremely dark regions, the function slope is extremely high, generating very high nonlinear gain, thus stretching weak signals submerged in background noise to the visually sensitive area; while in... The larger highlight areas have a gentler slope, effectively suppressing overexposure caused by overcompensation.

[0051] For illuminance constraint parameters, implementers can use a standard multi-level grayscale test card to take pictures in a standard darkroom environment to obtain the actual output brightness value of each grayscale test block. Combined with the standard reflectance corresponding to each grayscale test block, a photoelectric conversion characteristic curve is plotted. This curve is then fitted to obtain a power exponent for setting the illuminance constraint parameters. It is important to emphasize that since the dynamic compensation factor maps relative illumination intensity using a reciprocal power, to ensure that this nonlinear mapping curve exhibits convex function characteristics, thereby producing a substantial stretching and enhancement effect on weak signals in dark areas, the illuminance constraint parameter must be strictly greater than 1. If, due to the physical characteristics or calibration errors of a specific sensor, the fitted power exponent is not greater than 1, the mapping relationship will degenerate into linearity or even cause reverse compression of the dark area signal. Therefore, the fitted power exponent is compared with a preset empirical lower limit value that is strictly greater than 1, and the maximum value is taken. This final value is then set as the illuminance constraint parameter. This embodiment integrates numerous comparative calibration experiments in extremely dark monitoring scenarios, and combines the nonlinear perception characteristics of the human visual system with the Gamma response coefficient of general standard display devices, setting the empirical lower limit value to 2.2.

[0052] In complex real-world surveillance scenarios, subtle reflective textures are easily masked by ambient noise, making it difficult to identify small local targets. This invention extracts the maximum value of the target illumination component of all pixels in the image to be processed as a global normalization benchmark. This benchmark can adaptively adjust according to the overall lighting conditions of the scene: when the image is in an extremely dark environment, the extracted maximum value is small, and using it as the denominator can significantly and adaptively increase the global dynamic compensation base, allowing the weak reflective signals in dark areas to receive high-intensity nonlinear stretching, thus improving the contrast of small local targets; when there is a strong light source in the image, the extracted maximum value is large, and using it as the denominator can effectively suppress the global compensation amplitude, preventing overexposure in bright areas. Under the dynamic control of this global benchmark, nonlinear mapping is performed in combination with the relative intensity of the target illumination component of each pixel, ensuring the restoration of details in dark areas while preventing overexposure in highlights.

[0053] For example, Figure 7 This is a schematic diagram illustrating the effect of the dynamic compensation factor. Figure 7 As can be seen, the dynamic compensation factor achieves a differentiated numerical response distribution based on the local relative intensity, providing a physical modulation basis for dark area brightening and highlight suppression.

[0054] Furthermore, based on the initial pixel values ​​of each channel, the target illumination component, the dynamic compensation factor, and the dark field basis bias constant, the final enhanced image at the current time is calculated: ; in, Indicates the color channel at the current moment. Spatial x-coordinate with spatial ordinate Enhanced pixel values ​​at the location; Indicates the current time; Represents the spatial x-coordinate at the current time. with spatial ordinate Initial pixel value of the channel at that location; Indicates the color channel number, including the red, green, and blue channels; Represents the spatial x-coordinate at the current time. with spatial ordinate The target illumination component at the location; This represents the dark field basis bias constant. ; Represents the spatial x-coordinate at the current time. with spatial ordinate Dynamic compensation factor at the location; This represents the maximum value of the target illumination component for all pixels in the image to be processed at the current moment.

[0055] When the target illumination component is small, it indicates that the pixel is located in a deep shadow area of ​​the scene. If the initial pixel value of the channel is relatively large at this time, it indicates that there is a real physical target with high reflectivity at this location, such as the texture of an object hidden in the dark, rather than pure dark field noise. In this case, combined with a large dynamic compensation factor, the enhanced pixel value is significantly increased. In extremely low-light monitoring scenarios, the original brightness of the image is extremely low. If the initial pixel value of the channel is directly divided by the target illumination component which is close to 0, it will not only easily cause numerical calculation abnormalities, but also cause the dark current noise at the sensor's underlying layer to be infinitely amplified. Therefore, this invention introduces a dark field substrate bias constant, which is added to the target illumination component and used as a divisor to stably extract the real physical reflection characteristics of the scene. The extracted physical reflection characteristics are remodulated and mapped using the maximum value of the dynamic compensation factor and the target illumination component, so that the weak but real physical texture in the dark area obtains targeted high-intensity gain, thereby achieving adaptive enhancement of low-light contrast and effective suppression of pure dark field dynamic random noise.

[0056] Furthermore, a final enhanced image is generated based on the enhanced pixel values ​​of each color channel, and the final enhanced image is output to a television camera for presentation.

[0057] For example, Figure 8 This is a schematic diagram illustrating the final image enhancement effect. It shows that the tiny dynamic targets and faint reflective textures, originally submerged in dark background noise, are now... Figure 8 The image achieves high-intensity adaptive stretching and clear rendering, with a significant improvement in overall visual contrast and detail recognition, while avoiding overexposure of highlights.

Claims

1. A method for image enhancement of a television camera in low light environments, characterized in that, include: S1: Obtain the initial pixel values ​​of each color channel in the current and previous time steps of the image sequence to be processed; S2: Based on the initial pixel values ​​of each color channel, extract the initial illumination components of the current moment and the previous moment, and combine spatial structure fluctuations and temporal differences to construct the basic spatiotemporal fluctuation value of the current moment. Using the mean of the initial illumination components in the neighborhood as the basis of the signal intensity, the basic spatiotemporal fluctuation value is modulated to obtain the spatiotemporal illuminance fluctuation intensity at the current moment. S3: Construct an adaptive diffusion coefficient based on the global discrete characteristics of spatiotemporal illuminance fluctuation intensity, and use the adaptive diffusion coefficient to dynamically weight the initial illumination component to obtain the target illumination component at the current moment. S4: Construct a dynamic compensation factor using the relative intensity and global distribution extreme value of the target illumination component, and combine the initial pixel value of the channel, the target illumination component, and the dynamic compensation factor to generate the final enhanced image; The extraction of the initial illumination components at the current and previous times includes: The maximum value among the initial pixel values ​​of each color channel at the same spatial coordinate at the current time and the previous time is used as the initial illumination component at the corresponding spatial coordinate. The method for obtaining the spatial structure fluctuations is as follows: The mean of the absolute values ​​of the differences between the initial illumination component at the current moment and the initial illumination components in the defined spatial neighborhood is calculated and used as the spatial structure fluctuation value. The construction of the basic spatiotemporal fluctuation value at the current moment includes: The absolute value of the time difference between the initial illumination component at the current moment and the previous moment is added to the spatial structure fluctuation value to obtain the basic spatiotemporal fluctuation value at the current moment. The spatiotemporal illuminance fluctuation intensity satisfies the expression: ； in, Represents the spatial x-coordinate at the current time. with spatial ordinate The intensity of spatiotemporal illuminance fluctuations at a given location; Indicates the current moment; Represents the spatial x-coordinate at the current time. with spatial ordinate The fundamental spatiotemporal fluctuation value at that location; Representing the current time in terms of spatial x-axis with spatial ordinate The mean of the initial illumination components within the 8-neighborhood of the center; This represents the dark field basis bias constant. The dark field substrate bias constant is the average value of the output pixel values ​​of each color channel of an image when multiple frames are captured in a completely dark environment. The adaptive diffusion coefficient satisfies the expression: ； In the formula, Represents the spatial x-coordinate at the current time. with spatial ordinate The adaptive diffusion coefficient at the location; Indicates the current moment; This represents the standard deviation of the spatiotemporal illuminance fluctuation intensity of all pixels in the image to be processed at the current moment. Represents an exponential function with the natural constant as its base; The target illumination component satisfies the expression: ； in, Represents the spatial x-coordinate at the current time. with spatial ordinate The target illumination component at the location; Indicates the current moment; Represents the spatial x-coordinate at the current time. with spatial ordinate The initial illumination component at the location; The dynamic compensation factor satisfies the expression: ； In the formula, Represents the spatial x-coordinate at the current time. with spatial ordinate Dynamic compensation factor at the location; Indicates the current moment; This represents the maximum value of the target illumination component of all pixels in the image to be processed at the current moment; Indicates the illuminance constraint parameter. ; The illuminance constraint parameters are obtained by taking pictures of each grayscale test block in a standard darkroom environment using a standard multi-level grayscale test card. The actual output brightness value of each grayscale test block is obtained, and the photoelectric conversion characteristic curve is plotted by combining the standard reflectance corresponding to each grayscale test block. The power exponent is obtained by fitting the photoelectric conversion characteristic curve.

2. The image enhancement method for television cameras in low-light environments according to claim 1, characterized in that, The generation of the final enhanced image includes: the enhanced pixel values ​​of each color channel satisfying the expression: ;in, Indicates the color channel at the current moment. Spatial x-coordinate with spatial ordinate Enhanced pixel values ​​at the location; Indicates the current moment; Represents the spatial x-coordinate at the current time. with spatial ordinate Initial pixel value of the channel at that location; Indicates the color channel number; Represents the spatial x-coordinate at the current time. with spatial ordinate The target illumination component at the location; This represents the dark field basis bias constant; Represents the spatial x-coordinate at the current time. with spatial ordinate Dynamic compensation factor at the location; This represents the maximum value of the target illumination component of all pixels in the image to be processed at the current moment; the final enhanced image is generated based on the enhanced pixel values ​​of each color channel.

3. The image enhancement method for television cameras in low-light environments according to claim 1, characterized in that, The method for obtaining the illuminance constraint parameters includes: The actual output brightness values ​​of each grayscale test block were obtained by taking pictures in a standard darkroom environment using a standard multi-level grayscale test card. The photoelectric conversion characteristic curve was plotted by combining the standard reflectance corresponding to each grayscale test block, and the power exponent was obtained by fitting the photoelectric conversion characteristic curve. This power exponent was set as the illuminance constraint parameter.

Images