Low-illumination image quality adaptive enhancement method and system for video monitoring

Abstract

The embodiment of the invention provides a video monitoring-oriented low-illumination image quality adaptive enhancement method and system, and belongs to the field of image processing. The method comprises the following steps: determining an initial monitoring video, and obtaining a key frame image corresponding to the initial monitoring video; performing deblurring processing on the key frame image by using entropy information corresponding to the key frame image and introducing an illumination evaluation parameter to obtain a deblurred image corresponding to the key frame image; performing uncertainty analysis on the deblurred image to obtain a target blurred value corresponding to the deblurred image, and performing image adjustment on the deblurred image according to the target blurred value to obtain a corresponding uncertain image; performing low-light image enhancement processing on the uncertain image to obtain a target enhanced image; carrying out image edge monitoring on the target enhanced image to obtain a target edge point and target accuracy corresponding to the target edge point; and performing image enhancement adjustment on the target enhanced image according to the target accuracy until the target enhanced image meeting the preset accuracy is obtained.

Description

Technical Field

[0001] This invention relates to the field of image processing, and more particularly to a method and system for adaptive image quality enhancement in low light conditions for video surveillance. Background Technology

[0002] With the advancement of artificial intelligence technology and the deepening integration of industry, academia, research, and education, AI is increasingly being applied to drive the development of vocational education. To ensure teaching quality, vocational education typically uses video surveillance to record students' learning processes and teachers' teaching activities, and then uses video analysis to evaluate teaching effectiveness and learning outcomes. However, in the video acquisition stage, inconsistent ambient lighting conditions lead to varying video quality, which reduces the accuracy of video analysis and thus affects the reliability of vocational education evaluation and management. Summary of the Invention

[0003] The main objective of this invention is to provide a low-light image quality adaptive enhancement method and system for video surveillance, aiming to solve the problem in the prior art where inconsistent lighting conditions in the acquired surveillance video environment lead to inconsistent video quality, which in turn reduces the accuracy of video analysis and affects the reliability of vocational education evaluation and management.

[0004] In a first aspect, embodiments of the present invention provide a low-light image quality adaptive enhancement method for video surveillance, comprising: Determine the initial monitoring video and obtain the keyframe images corresponding to the initial monitoring video; The keyframe image is deblurred by using the entropy information corresponding to the keyframe image and introducing illumination evaluation parameters to obtain the deblurred image corresponding to the keyframe image. Uncertainty analysis is performed on the deblurred image to obtain the target blur value corresponding to the deblurred image, and image adjustment is performed on the deblurred image according to the target blur value to obtain the corresponding uncertain image; The uncertain image is subjected to low-light image enhancement processing to obtain a target enhanced image; and the target enhanced image is subjected to image edge detection to obtain target edge points and the target accuracy corresponding to the target edge points; The target enhanced image is adjusted according to the target accuracy until the target enhanced image that meets the preset accuracy is obtained.

[0005] Secondly, embodiments of the present invention provide a low-light image quality adaptive enhancement system for video surveillance, comprising: The image acquisition module is used to determine the initial monitoring video and obtain the keyframe image corresponding to the initial monitoring video; The image processing module is used to deblur the keyframe image by using the entropy information corresponding to the keyframe image and introducing illumination evaluation parameters, so as to obtain the deblurred image corresponding to the keyframe image. An image adjustment module is used to perform uncertainty analysis on the deblurred image to obtain a target blur value corresponding to the deblurred image, and to perform image adjustment on the deblurred image according to the target blur value to obtain a corresponding uncertain image; An edge detection module is used to perform low-light image enhancement processing on the uncertain image to obtain a target enhanced image; and to perform image edge detection on the target enhanced image to obtain target edge points and the target accuracy corresponding to the target edge points; An enhancement adjustment module is used to perform image enhancement adjustment on the target enhanced image according to the target accuracy until the target enhanced image that meets the preset accuracy is obtained.

[0006] This invention provides a low-light image quality adaptive enhancement method and system for video surveillance. The method includes: determining an initial surveillance video and obtaining keyframe images corresponding to the initial surveillance video; using entropy information corresponding to the keyframe images and introducing illumination evaluation parameters to deblur the keyframe images to obtain a deblurred image corresponding to the keyframe images, thereby specifically improving the blurring caused by uneven illumination or motion while preserving key information. The introduced low-light image enhancement processing directly addresses the challenge of uneven lighting, significantly improving image visibility and detail in low-light scenes; performing uncertainty analysis on the deblurred image to obtain a target blur value corresponding to the deblurred image, and adjusting the deblurred image according to the target blur value to obtain a corresponding uncertain image; performing low-light image enhancement processing on the uncertain image to obtain a target enhanced image; and performing image edge detection on the target enhanced image to obtain target edge points and the target accuracy corresponding to the target edge points; and adjusting the target enhanced image according to the target accuracy until a target enhanced image that meets a preset accuracy is obtained. Thus, through uncertainty analysis and iterative adjustment based on edge accuracy, a quality assurance closed loop is formed. It can automatically determine whether the processing effect is good enough and continuously optimize it until it outputs clear, well-defined, high-quality images suitable for machine analysis, thus fundamentally ensuring the reliability of teaching quality and learning outcome assessments based on these videos. This method not only solves specific image quality problems such as uneven lighting, blurriness, and low light, but also ensures stable, high-quality images for analysis even in complex real-world environments by establishing an automated evaluation-feedback-iteration mechanism, making AI-driven vocational education assessment systems more reliable and effective. Attached Figure Description

[0007] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0008] Figure 1 This is a flowchart illustrating a low-light image quality adaptive enhancement method for video surveillance provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the module structure of a low-light image quality adaptive enhancement system for video surveillance, provided in an embodiment of the present invention. Detailed Implementation

[0009] 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.

[0010] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content and operations / steps, nor does it necessarily have to be performed in the described order. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.

[0011] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0012] This invention provides a low-light image quality adaptive enhancement method and system for video surveillance. The low-light image quality adaptive enhancement method for video surveillance can be applied to terminal devices, such as tablets, laptops, desktop computers, personal digital assistants, and wearable devices. The terminal device can be a server or a server cluster.

[0013] The following detailed description of some embodiments of the present invention is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0014] Please refer to Figure 1 , Figure 1This is a flowchart illustrating a low-light image quality adaptive enhancement method for video surveillance, provided in an embodiment of the present invention.

[0015] like Figure 1 As shown, the low-light image quality adaptive enhancement method for video surveillance includes steps S101 to S105.

[0016] Step S101: Determine the initial monitoring video and obtain the keyframe image corresponding to the initial monitoring video.

[0017] For example, initial monitoring video is obtained by using cameras in the target classroom or training site. Then, the visual differences between adjacent video frames are continuously calculated, such as by comparing changes in the pixel histogram. When the difference exceeds a preset threshold, it means that the scene or action has changed significantly. At this point, the current frame is identified and extracted as a keyframe, thereby obtaining all the keyframe images corresponding to the initial monitoring video.

[0018] Step S102: Use the entropy information corresponding to the keyframe image and introduce illumination evaluation parameters to perform deblurring on the keyframe image to obtain the deblurred image corresponding to the keyframe image.

[0019] For example, information entropy is calculated for the keyframe image. Entropy quantifies the degree of disorder or information richness in the gray-level distribution of an image. Regions with higher entropy values ​​typically contain rich details and textures, while regions with lower entropy values ​​may be smoother. Simultaneously, the lighting evaluation parameters of the keyframe image are analyzed. This typically includes metrics such as global brightness, contrast, and lighting uniformity, used to objectively determine the lighting conditions at the time of capture.

[0020] For example, entropy information is correlated with illumination evaluation parameters. Under low illumination conditions, image blurring may mainly be due to noise amplification and loss of detail, and the deblurring strategy is noise reduction and signal enhancement. In areas of high brightness or drastic illumination changes, blurring may be due to local overexposure or motion blur, and the deblurring strategy is dynamic range restoration and motion compensation. High entropy regions are given higher restoration priority and more refined processing intensity.

[0021] For example, the parameters of the deblurring algorithm are adaptively selected or adjusted for different regions of the keyframe image. For low-light, high-entropy regions, a deblurring algorithm with stronger noise robustness, such as iterative restoration based on a prior model, is used, while strictly controlling noise amplification. For regions dominated by motion blur, the direction and size of the blur kernel (point spread function) are estimated, and a deconvolution operation is performed. The key parameters are dynamically adjusted according to the entropy information and lighting conditions.

[0022] For example, the adjusted algorithm model is used to process the entire keyframe image. The final output is a deblurred image.

[0023] In some embodiments, the step of deblurring the keyframe image using entropy information corresponding to the keyframe image and introducing illumination evaluation parameters to obtain a deblurred image corresponding to the keyframe image includes: performing entropy information analysis on the keyframe image to obtain the target brightness region corresponding to the keyframe image; mapping the keyframe image to a target coordinate system centered on the target brightness region to obtain target coordinate information of the keyframe image in the target coordinate system; calculating distance information between each first sub-pixel in the keyframe image and the target brightness region, and determining the initial irradiance corresponding to the first sub-pixel based on the distance information; performing data clustering on the keyframe image based on the initial irradiance and the target coordinate information to obtain target clusters; calculating reliability parameters for each sub-cluster in the target cluster to obtain the illumination evaluation parameters corresponding to the sub-cluster; optimizing the initial irradiance in the sub-cluster based on the illumination evaluation parameters to obtain the target irradiance corresponding to the sub-cluster; and performing deblurring on the keyframe image based on the target irradiance and the target brightness region to obtain the deblurred image corresponding to the keyframe image.

[0024] For example, block entropy analysis is performed on the keyframe image, dividing the keyframe image into multiple local windows, calculating the information entropy of each window, and statistically analyzing the average brightness. Regions with high entropy values ​​usually contain rich edge and texture information and are the focus of priority for deblurring while preserving fidelity. Combining entropy value and brightness distribution, a connected region with high entropy value and moderate brightness is selected as the target brightness region, which will be regarded as the center of subsequent spatial reference and lighting reference.

[0025] For example, a new coordinate system (such as a polar coordinate system or a Cartesian offset coordinate system) is established with the geometric center of the target brightness region as the origin. The coordinates of all pixels in the original image are transformed into this coordinate system to obtain the position vector (distance and azimuth, or horizontal and vertical offset) of each pixel relative to the center, i.e., the target coordinate information. This step converts the absolute position into a relative position with the illumination center as the reference, providing a unified spatial reference for subsequent distance measurement and clustering.

[0026] For example, the spatial distance from each first sub-pixel in the image to the center of the target brightness region is calculated. Based on a physical model of illumination attenuation (such as the inverse square decay of point light source radiance with distance, or a more general exponential decay function), this distance is mapped to an initial irradiance value. The closer the distance, the higher the irradiance value is assigned; conversely, the farther the distance, the lower the irradiance value. This yields the initial irradiance corresponding to each first sub-pixel in the entire image.

[0027] For example, the target coordinate information of each first sub-pixel is combined with the initial irradiance to form a two-dimensional or three-dimensional feature vector, and an unsupervised clustering algorithm such as K-means is used to perform spatial radiative joint clustering to obtain the target clustering result.

[0028] For example, a reliability analysis is performed on each sub-cluster in the target clustering result. For instance, the variance or standard deviation of the initial irradiance within each sub-cluster is calculated. The smaller the value, the more stable the illumination estimation in that area. The average distance from the pixel to the cluster center within the sub-cluster reflects the spatial continuity of the area, and the average gradient magnitude at the boundary of the sub-cluster reflects whether the area is a texture-rich region. Thus, the above indicators are integrated into illumination evaluation parameters to quantify the reliability of the illumination estimation or the quality of the illumination conditions for that sub-cluster.

[0029] For example, the initial irradiance is adaptively optimized using the illumination evaluation parameters of each sub-cluster. For sub-clusters with high reliability, their initial irradiance is directly retained or only slightly smoothed; for sub-clusters with low reliability, weighted guided filtering or irradiance interpolation from adjacent reliable clusters is used to correct their irradiance values ​​and suppress outlier estimations. This ultimately yields a smoother target irradiance for each pixel.

[0030] For example, the optimized target irradiance and target brightness region are used together as prior information to construct a deblurring model. For instance, the image is modeled as the product of the irradiance component and the reflectance component. The blur kernel is estimated using the edge information of the irradiance map, and then non-blind deconvolution is performed to output a deblurred image that has been sharpened and has strong illumination consistency, providing high-quality input for subsequent uncertainty analysis and enhancement adjustment.

[0031] In some embodiments, the step of performing entropy information analysis on the keyframe image to obtain the target brightness region corresponding to the keyframe image includes: performing target recognition on the keyframe image to obtain target structure information corresponding to the keyframe image; performing brightness analysis on the target structure information to obtain row brightness information and column brightness information corresponding to the target structure information; performing neighborhood analysis based on the row brightness information and the column brightness information to obtain an initial brightness region; and performing entropy information analysis on the initial brightness region to obtain the target brightness region.

[0032] For example, deep learning-based object detection models such as YOLO perform forward inference on keyframe images. These models, specifically trained for vocational education scenarios, can identify structural teaching elements in keyframe images, such as blackboard / whiteboard writing areas, lectern desktops, practical training workstations, teacher and student faces, or textbook screens. The output is a pixel-level mask or bounding box for each object instance; these masks and bounding boxes together constitute the object's structural information. Irrelevant backgrounds such as walls and floors are filtered out, focusing the analysis on the core area where teaching activities occur.

[0033] For example, pixel-by-pixel brightness is extracted for each target structural region, and cumulative brightness is calculated along the horizontal (row) and vertical (column) directions respectively. The brightness values ​​of all pixels in each row are summed or averaged to obtain a one-dimensional curve with the row number as the x-axis and the brightness value as the y-axis. Similarly, brightness accumulation is performed on each column to obtain another one-dimensional curve. These two curves intuitively reflect the brightness distribution trend within the target region: the peak area of ​​the row brightness curve corresponds to the horizontal band with strong illumination, and the peak area of ​​the column brightness curve corresponds to the vertical band with strong illumination. The row brightness curve is smoothed and local peak detection is performed to extract the row intervals with significantly higher brightness than the neighborhood; the same operation is performed on the column brightness curve to extract the column intervals with significantly higher brightness than the neighborhood. The Cartesian product of the bright row intervals and the bright column intervals is taken to obtain several rectangular candidate boxes (i.e., intersection areas with higher brightness). Subsequently, using the pixels in these rectangular boxes as seeds, four-connected or eight-connected region growth is performed within the target structural region to aggregate all connected pixels with brightness values ​​higher than the adaptive threshold to form the initial brightness region. The initial brightness area is spatially continuous, and the overall brightness level is significantly higher than that of the surrounding area.

[0034] For example, the entropy of each initial brightness region is calculated to obtain entropy value information, and then the region with the largest entropy value is selected as the target brightness region from all initial brightness regions. This decision ensures that the selected region is not only the location with the strongest physical illumination, but also the key part where the teaching content information is most concentrated and needs to be clearly restored, thus providing an optimal reference benchmark for subsequent irradiance modeling and deblurring processing.

[0035] In some embodiments, optimizing the initial irradiance in the sub-cluster based on the illumination evaluation parameters to obtain the target irradiance corresponding to the sub-cluster includes: determining preset evaluation parameters, and determining, based on the illumination evaluation parameters and the preset evaluation parameters, the irradiance to be optimized and the first irradiance not to be optimized in the sub-cluster; extracting edge features from the keyframe image to obtain an edge feature map corresponding to the keyframe image; generating a target guidance image based on the edge feature map, and filtering the irradiance to be optimized based on the target guidance image to obtain a second irradiance corresponding to the irradiance to be optimized; and determining the target irradiance corresponding to the sub-cluster based on the first irradiance and the second irradiance.

[0036] For example, based on the comparison between the illumination evaluation parameters of the sub-cluster and the preset evaluation parameters, the irradiance within the sub-cluster is classified as the first irradiance and directly retained, while unreliable irradiance is classified as irradiance to be optimized and then enters the subsequent correction process.

[0037] For example, edge feature maps are obtained by extracting edge features from keyframe images using gradient operators or deep learning edge detection models.

[0038] For example, using the edge feature map as a structural prior, a target guiding image is generated. Guided filtering or weighted least squares filtering is then performed on the region where unreliable irradiance is located, so that the corrected irradiance maintains sharp edges while achieving a smooth transition in flat areas, thereby obtaining a filtered image. The irradiance is then recalculated on the filtered image to obtain the second irradiance after optimization.

[0039] For example, the retained first irradiance and the filtered second irradiance are fused at their respective sub-cluster locations to form a final consistent and physically reasonable target irradiance, providing a robust illumination estimation benchmark for subsequent deblurring processing.

[0040] Step S103: Perform uncertainty analysis on the deblurred image to obtain the target blur value corresponding to the deblurred image, and perform image adjustment on the deblurred image according to the target blur value to obtain the corresponding uncertain image.

[0041] For example, the deblurred image is deconvolved back to the blurred state, and the difference is calculated with the keyframe image. Areas with large differences usually indicate that the algorithm has over-sharpened or produced artifacts, resulting in high uncertainty. Alternatively, the deblurred image can be divided into a grid, and each grid can be quantized based on the difference, thereby taking the weighted average of all grid scores, or the score of the worst region can be directly taken as the target blur value for the entire image.

[0042] For example, the higher the target blur value, the less reliable the deblurring result, and the greater the subsequent adjustment. If the target blur value is high, a larger radius is chosen for Gaussian blurring; if the value is low, only slight texture smoothing is performed, resulting in an uncertain image.

[0043] In some implementations, the step of performing uncertainty analysis on the deblurred image to obtain the target blur value corresponding to the deblurred image includes: normalizing the deblurred image to obtain a normalized image; determining blur control parameters, and calculating a membership function for each second sub-pixel in the normalized image according to the blur control parameters to obtain a first data value corresponding to the second sub-pixel; calculating a non-membership function for each second sub-pixel in the normalized image according to the blur control parameters to obtain a second data value corresponding to the second sub-pixel; and fusing the first data value and the second data value to determine the target blur value corresponding to the second sub-pixel; wherein the first data value and the second data value are obtained according to the following formula: ; ; in, This represents the first data value corresponding to the second sub-pixel when the horizontal position is i and the vertical position is j; The fuzzy control parameters are represented by c, d, and f; c, d, and f represent constants. This represents the pixel value of the second sub-pixel when its horizontal position is i and its vertical position is j, as shown in the normalized image. This represents the second data value corresponding to the second sub-pixel when the horizontal position is i and the vertical position is j.

[0044] For example, the pixel values ​​of the deblurred image are uniformly mapped to the [0,1] interval to eliminate the influence of dimensions and make it suitable for subsequent membership calculation.

[0045] For example, fuzzy control parameters are set according to actual needs or historical experience, and the membership function of each second sub-pixel in the normalized image is calculated according to the following formula to obtain the first data value corresponding to the second sub-pixel: ; in, This represents the first data value corresponding to the second sub-pixel when the horizontal position is i and the vertical position is j; The fuzzy control parameters are represented by c; c represents a constant. This represents the pixel value of the second sub-pixel in the normalized image when its horizontal position is i and its vertical position is j.

[0046] For example, the second data value corresponding to each second sub-pixel in the normalized image is obtained by performing a non-membership function calculation based on the fuzzy control parameters using the following formula: ; in, The fuzzy control parameters are represented by d and f, which are constants. This represents the pixel value of the second sub-pixel when its horizontal position is i and its vertical position is j, as shown in the normalized image. This represents the second data value corresponding to the second sub-pixel when the horizontal position is i and the vertical position is j.

[0047] For example, the first data value and the second data value of each second sub-pixel are fused to obtain the degree of uncertainty of the pixel, i.e., the target blur value.

[0048] In some implementations, the step of fusing the first data value and the second data value to determine the target blur value corresponding to the second sub-pixel includes: performing a cube calculation on the first data value to obtain a first value, and performing a cube calculation on the second data value to obtain a second value; calculating the sum between the first value and the second value to obtain a target value, and determining the target blur value corresponding to the second sub-pixel based on the target value.

[0049] For example, the first data value is cubed to obtain the first value, and then the second data value is cubed in the same way to obtain the second value.

[0050] For example, the first and second values ​​of the same pixel are added together to obtain the target value, and then the target value is obtained by subtracting the target value from the constant 1 and taking the cube root.

[0051] Step S104: Perform low-light image enhancement processing on the uncertain image to obtain a target enhanced image; and perform image edge detection on the target enhanced image to obtain target edge points and the target accuracy corresponding to the target edge points.

[0052] For example, when performing low-light enhancement on an uncertain image, the illumination component of the image is extracted by multi-scale Gaussian filtering or guided filtering. The illumination component is subtracted from the original image in the logarithmic domain to obtain the reflection component. The illumination component is then subjected to adaptive gamma correction to enhance the brightness of dark areas. Finally, the corrected illumination component is multiplied by the reflection component to obtain the target enhanced image.

[0053] For example, edge detection is performed on the target enhancement image. First, it is converted into a grayscale image and Gaussian smoothed. The gradient magnitude and direction are calculated using the Sobel or Scharr operator. The edges are thinned by non-maximum suppression. Then, edge connection and filtering are performed based on the double threshold automatically set by the gradient magnitude histogram. Finally, all target edge points are extracted, and the gradient magnitude of each edge point is normalized to the range of 0 to 1 as its target accuracy. The higher the accuracy, the more significant the edge and the more reliable the detection result.

[0054] In some embodiments, the step of performing low-light image enhancement processing on the uncertain image to obtain a target enhanced image includes: performing wavelet transform processing on the uncertain image to obtain initial low-frequency wavelet coefficients corresponding to the uncertain image; determining a filtering threshold and performing range reduction processing on the initial high-frequency wavelet coefficients according to the filtering threshold to obtain target high-frequency wavelet coefficients; performing image enhancement processing on the uncertain image according to the target high-frequency wavelet coefficients to obtain a first image; performing blurring processing on the initial low-frequency wavelet coefficients to obtain target low-frequency wavelet coefficients; performing image enhancement processing on the uncertain image according to the target low-frequency wavelet coefficients to obtain a second image; and performing image fusion on the first image and the second image to obtain the target enhanced image.

[0055] For example, an uncertain image is decomposed into a low-frequency subband and a high-frequency subband by performing a discrete wavelet transform, thereby obtaining low-frequency wavelet coefficients and high-frequency wavelet coefficients.

[0056] For example, threshold screening is performed on the initial high-frequency wavelet coefficients to suppress noise and enhance effective details. For instance, the initial high-frequency wavelet coefficients can be shrunk or truncated based on the standard deviation estimation of the wavelet coefficients to obtain the denoised and edge-sharpened target high-frequency wavelet coefficients, preserving and highlighting important details.

[0057] For example, the first image is reconstructed by performing an inverse wavelet transform using the initial low-frequency wavelet coefficients and the target high-frequency wavelet coefficients.

[0058] For example, the initial low-frequency wavelet coefficients are blurred to extract the illumination components of the scene and eliminate local brightness unevenness. For instance, a Gaussian low-pass filter is applied to the low-frequency sub-band image in the spatial domain, with the standard deviation set according to the image size and the degree of illumination variation, thereby obtaining the smoothed target low-frequency wavelet coefficients.

[0059] For example, an inverse wavelet transform is performed using the target low-frequency wavelet coefficients and the initial high-frequency wavelet coefficients to reconstruct the second image. This allows the first and second images to be fused, balancing detail and brightness, resulting in a fused enhanced image that retains rich detail and a uniform, natural brightness distribution even in low-light conditions.

[0060] In some embodiments, the step of performing image enhancement processing on the uncertain image based on the target low-frequency wavelet coefficients to obtain a second image includes: determining the minimum and maximum pixel values ​​corresponding to the uncertain image, and determining the pixel membership degree corresponding to each pixel position in the uncertain image based on the minimum and maximum pixel values; obtaining the relevant pixels corresponding to each pixel position under the target window from the uncertain image, and calculating the mean membership degree corresponding to the pixel position based on the relevant pixels; determining an adjustment factor, and determining the pixel contrast corresponding to the pixel position based on the adjustment factor using the pixel membership degree and the mean membership degree; performing blur enhancement on the pixel position based on the pixel contrast and the mean membership degree to obtain the enhanced membership degree corresponding to the pixel position; determining the enhanced pixel value corresponding to the pixel position based on the maximum and minimum pixel values ​​combined with the enhanced membership degree; and obtaining the second image based on the enhanced pixel value and the pixel position.

[0061] For example, the entire uncertain image is traversed, all pixel grayscale values ​​are counted, and the minimum and maximum pixel values ​​are found. For each pixel location in the uncertain image, the pixel membership degree corresponding to the pixel location is determined by combining the minimum and maximum pixel values.

[0062] For example, a fixed-size neighborhood window, such as 3x3, is selected with the pixel position as the center to obtain the target window. Then, the membership degree of the relevant pixels under the target window is calculated, and the mean membership degree is calculated to obtain the mean membership degree corresponding to the target window. The mean membership degree reflects the overall brightness of the neighborhood of the pixel position and is used to measure the contrast of the center pixel relative to the local area.

[0063] For example, pixel contrast characterizes the degree of difference between the pixel membership degree and the local mean membership degree. An adjustment factor is determined empirically, and then the absolute value of the difference between the pixel membership degree and the mean membership degree is calculated. Then, the absolute value of the sum of the square of the adjustment factor and the mean membership degree is calculated. Finally, the pixel contrast is obtained by dividing the absolute value of the difference by the absolute value of the sum.

[0064] For example, the pixel membership corresponding to the pixel position is enhanced by a non-linear transformation based on the pixel contrast and mean membership, thereby obtaining the enhanced membership corresponding to the pixel position.

[0065] For example, the enhanced membership is mapped back to the color space using the minimum and maximum pixel values, thereby generating new pixel values, and all the new pixels make up the second image.

[0066] In some embodiments, the step of performing blur enhancement on the pixel position based on the pixel contrast and the mean membership degree to obtain the enhanced membership degree corresponding to the pixel position includes: comparing the mean membership degree and the pixel membership degree to obtain a target comparison result; when the target comparison result satisfies a first preset result, calculating the difference between the pixel contrast and a preset constant to obtain a first difference and calculating the sum between the pixel contrast and the preset constant to obtain a target sum; calculating a first product between the mean membership degree and the first difference, and calculating the division result between the first product and the target sum to obtain the pixel position. The corresponding enhanced membership degree; when the target comparison result satisfies the second preset result, the difference between the pixel contrast and the preset constant is calculated to obtain the first difference value, and the sum of the pixel contrast and the preset constant is calculated to obtain the target sum value; the difference between the mean membership degree and the preset constant is calculated to obtain the second difference value; the product between the first difference value and the second difference value is calculated to obtain the second product, and the difference between the preset constant value and the second product value is calculated to obtain the third difference value; the enhanced membership degree corresponding to the pixel position is obtained by performing a division operation based on the third difference value and the target sum value.

[0067] For example, the preset constant is a fixed value, usually 1, but can also be adjusted according to the characteristics of the image.

[0068] For example, the mean membership degree and the pixel membership degree are compared, and the target comparison result is that the mean membership degree is greater than or equal to the pixel membership degree, or the mean membership degree is less than either of the two.

[0069] For example, the first preset result is that the mean membership degree is greater than or equal to the pixel membership degree, and the second preset result is that the mean membership degree is less than the pixel membership degree.

[0070] For example, when the target comparison result is the first preset result, the difference between the pixel contrast and the preset constant is calculated to obtain the first difference, and the sum of the pixel contrast and the preset constant is calculated to obtain the target sum. Then, the first product between the mean membership degree and the first difference is calculated, and the division result between the first product and the target sum is calculated to obtain the enhanced membership degree corresponding to the pixel position.

[0071] For example, when the target comparison result is the second preset result, the difference between the pixel contrast and the preset constant is calculated to obtain the first difference, and the sum of the pixel contrast and the preset constant is calculated to obtain the target sum; the difference between the mean membership degree and the preset constant is calculated to obtain the second difference; the product between the first difference and the second difference is calculated to obtain the second product, and the difference between the preset constant and the second product is calculated to obtain the third difference; thereby, the enhanced membership degree corresponding to the pixel position is obtained by performing a division operation based on the third difference and the target sum.

[0072] Specifically, the algorithm compares the average brightness of a pixel with that of its neighborhood, selects different nonlinear transformation formulas, and selectively stretches the contrast in the fuzzy domain, thereby providing support for subsequent image enhancement.

[0073] Step S105: Adjust the target enhancement image according to the target accuracy until the target enhancement image that meets the preset accuracy is obtained.

[0074] For example, the target accuracy is compared with the preset accuracy to obtain a comparison result. When the comparison result is that the target accuracy is greater than or equal to the preset accuracy, the target enhanced image is directly obtained. If the comparison result is that the target accuracy is less than the preset accuracy, the parameters are adjusted and steps S102 to S104 are re-executed until the target enhanced image with a target accuracy greater than or equal to the preset accuracy is obtained.

[0075] Please see Figure 2 , Figure 2This application provides a low-light image quality adaptive enhancement system 200 for video surveillance. The system includes an image acquisition module 201, an image processing module 202, an image adjustment module 203, an edge detection module 204, and an enhancement adjustment module 205. The image acquisition module 201 determines an initial surveillance video and obtains a keyframe image corresponding to the initial surveillance video. The image processing module 202 uses entropy information corresponding to the keyframe image and incorporates illumination evaluation parameters to deblur the keyframe image, obtaining a deblurred image corresponding to the keyframe image. The image adjustment module 203 is used to perform uncertainty analysis on the deblurred image to obtain the target blur value corresponding to the deblurred image, and to perform image adjustment on the deblurred image according to the target blur value to obtain the corresponding uncertain image; the edge detection module 204 is used to perform low-light image enhancement processing on the uncertain image to obtain the target enhanced image; and to perform image edge detection on the target enhanced image to obtain the target edge point and the target accuracy corresponding to the target edge point; the enhancement adjustment module 205 is used to perform image enhancement adjustment on the target enhanced image according to the target accuracy until the target enhanced image that meets the preset accuracy is obtained.

[0076] In some implementations, the low-light image quality adaptive enhancement system 200 for video surveillance can be applied to terminal devices.

[0077] It should be noted that those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the low-light image quality adaptive enhancement system 200 for video surveillance described above can be referred to the corresponding process in the aforementioned embodiment of the low-light image quality adaptive enhancement method for video surveillance, and will not be repeated here.

[0078] This invention also provides a storage medium for computer-readable storage, wherein the storage medium stores one or more programs that can be executed by one or more processors to implement the steps of any of the low-light image quality adaptive enhancement methods for video surveillance provided in the specification of this invention.

[0079] The storage medium can be an internal storage unit of the terminal device described in the foregoing embodiments, such as the hard drive or memory of the terminal device. Alternatively, the storage medium can be an external storage device of the terminal device, such as a plug-in hard drive, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the terminal device.

[0080] Those skilled in the art will understand that all or some of the steps, systems, or apparatuses disclosed above, and their functional modules / units, can be implemented as software, firmware, hardware, or suitable combinations thereof. In hardware embodiments, the division between functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed collaboratively by several physical components. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and can be accessed by a computer. Furthermore, it is well known to those skilled in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0081] It should be understood that the term "and / or" as used in this specification and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes such combinations. It should be noted that, herein, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0082] The sequence numbers of the above embodiments of the present invention are merely for descriptive purposes and do not represent the superiority or inferiority of the embodiments. The above descriptions are only specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A low-light image quality adaptive enhancement method for video surveillance, characterized in that, The method includes: Determine the initial monitoring video and obtain the keyframe images corresponding to the initial monitoring video; The keyframe image is deblurred by using the entropy information corresponding to the keyframe image and introducing illumination evaluation parameters to obtain the deblurred image corresponding to the keyframe image. Uncertainty analysis is performed on the deblurred image to obtain the target blur value corresponding to the deblurred image, and image adjustment is performed on the deblurred image according to the target blur value to obtain the corresponding uncertain image; The uncertain image is subjected to low-light image enhancement processing to obtain a target enhanced image; and the target enhanced image is subjected to image edge detection to obtain target edge points and the target accuracy corresponding to the target edge points. The target enhanced image is adjusted according to the target accuracy until the target enhanced image that meets the preset accuracy is obtained.

2. The method according to claim 1, characterized in that, The step of using the entropy information corresponding to the keyframe image and introducing illumination evaluation parameters to perform deblurring on the keyframe image to obtain the deblurred image corresponding to the keyframe image includes: Entropy information analysis is performed on the keyframe image to obtain the target brightness region corresponding to the keyframe image; The keyframe image is mapped to a target coordinate system centered on the target brightness region to obtain the target coordinate information of the keyframe image in the target coordinate system; For each first sub-pixel in the keyframe image, calculate the distance information between it and the target brightness region, and determine the initial irradiance corresponding to the first sub-pixel based on the distance information; Based on the initial irradiance and the target coordinate information, the keyframe image is clustered to obtain target clusters; The illumination evaluation parameters corresponding to each sub-cluster in the target cluster are obtained by calculating the reliability parameters. The initial irradiance in the sub-cluster is optimized based on the light evaluation parameters to obtain the target irradiance corresponding to the sub-cluster; The keyframe image is deblurred based on the target irradiance and the target brightness region to obtain the deblurred image corresponding to the keyframe image.

3. The method according to claim 2, characterized in that, The step of performing entropy information analysis on the keyframe image to obtain the target brightness region corresponding to the keyframe image includes: Target recognition is performed on the keyframe image to obtain the target structure information corresponding to the keyframe image; Brightness analysis is performed on the target structure information to obtain the row brightness information and column brightness information corresponding to the target structure information; The initial brightness region is obtained by performing neighborhood analysis based on the row brightness information and the column brightness information; Entropy information analysis is performed on the initial brightness region to obtain the target brightness region.

4. The method according to claim 2, characterized in that, The step of optimizing the initial irradiance of the sub-cluster based on the illumination evaluation parameters to obtain the target irradiance corresponding to the sub-cluster includes: Determine preset evaluation parameters, and based on the illumination evaluation parameters and the preset evaluation parameters, determine the irradiance to be optimized and the first irradiance that does not need to be optimized in the sub-cluster; Edge feature extraction is performed on the keyframe image to obtain the edge feature map corresponding to the keyframe image; A target guidance image is generated based on the edge feature map, and the irradiance to be optimized is filtered based on the target guidance image to obtain a second irradiance corresponding to the irradiance to be optimized. The target irradiance corresponding to the sub-cluster is determined based on the first irradiance and the second irradiance.

5. The method according to claim 1, characterized in that, The step of performing uncertainty analysis on the deblurred image to obtain the target blur value corresponding to the deblurred image includes: The deblurred image is normalized to obtain a normalized image; Determine the fuzzy control parameters, and calculate the membership function for each second sub-pixel in the normalized image based on the fuzzy control parameters to obtain the first data value corresponding to the second sub-pixel; Based on the fuzzy control parameters, a non-membership function is calculated for each second sub-pixel in the normalized image to obtain the second data value corresponding to the second sub-pixel; The target blur value corresponding to the second sub-pixel is determined by fusing the first data value and the second data value; The first data value and the second data value are obtained according to the following formulas: ； ； in, This represents the first data value corresponding to the second sub-pixel when the horizontal position is i and the vertical position is j; The fuzzy control parameters are represented by c, d, and f; c, d, and f represent constants. This represents the pixel value of the second sub-pixel in the normalized image when its horizontal position is i and its vertical position is j. This represents the second data value corresponding to the second sub-pixel when the horizontal position is i and the vertical position is j.

6. The method according to claim 5, characterized in that, The step of fusing the first data value and the second data value to determine the target blur value corresponding to the second sub-pixel includes: The first data value is cubed to obtain a first value, and the second data value is cubed to obtain a second value; The target value is obtained by calculating the sum between the first value and the second value, and the target blur value corresponding to the second sub-pixel is determined based on the target value.

7. The method according to claim 1, characterized in that, The step of performing low-light image enhancement processing on the uncertain image to obtain the target enhanced image includes: The uncertain image is subjected to wavelet transform processing to obtain the initial low-frequency wavelet coefficients corresponding to the uncertain image; Determine the filtering threshold, and narrow down the range of the initial high-frequency wavelet coefficients according to the filtering threshold to obtain the target high-frequency wavelet coefficients; The uncertain image is enhanced using the target high-frequency wavelet coefficients to obtain a first image; The initial low-frequency wavelet coefficients are fuzzified to obtain the target low-frequency wavelet coefficients; The uncertain image is enhanced using the target low-frequency wavelet coefficients to obtain a second image. The target enhanced image is obtained by image fusion based on the first image and the second image.

8. The method according to claim 7, characterized in that, The step of performing image enhancement processing on the uncertain image based on the target low-frequency wavelet coefficients to obtain a second image includes: Determine the minimum and maximum pixel values ​​corresponding to the uncertain image, and determine the pixel membership degree corresponding to each pixel position in the uncertain image based on the minimum and maximum pixel values; Obtain the relevant pixels corresponding to each pixel position under the target window from the uncertain image, and calculate the mean membership degree corresponding to the pixel position based on the relevant pixels; Determine the adjustment factor, and determine the pixel contrast corresponding to the pixel position based on the adjustment factor using the pixel membership degree and the mean membership degree; The pixel position is blurred and enhanced based on the pixel contrast and the mean membership degree to obtain the enhanced membership degree corresponding to the pixel position; The enhanced pixel value corresponding to the pixel position is determined based on the maximum and minimum pixel values ​​combined with the enhanced membership degree. The second image is obtained based on the enhanced pixel value and the pixel position.

9. The method according to claim 8, characterized in that, The step of performing blur enhancement on the pixel position based on the pixel contrast and the mean membership degree to obtain the enhanced membership degree corresponding to the pixel position includes: By comparing the mean membership degree and the pixel membership degree, a target comparison result is obtained; When the target comparison result satisfies the first preset result, the difference between the pixel contrast and the preset constant is calculated to obtain the first difference, and the sum between the pixel contrast and the preset constant is calculated to obtain the target sum value. Calculate the first product between the mean membership degree and the first difference, and calculate the division result between the first product and the target sum value to obtain the enhanced membership degree corresponding to the pixel position; When the target comparison result satisfies the second preset result, the difference between the pixel contrast and the preset constant is calculated to obtain the first difference, and the summation between the pixel contrast and the preset constant is calculated to obtain the target sum value. The second difference is obtained by calculating the difference between the mean membership degree and the preset constant; The product of the first difference and the second difference is calculated to obtain the second product, and the difference between the preset constant and the second product is calculated to obtain the third difference; The enhanced membership degree corresponding to the pixel position is obtained by performing a division operation based on the third difference and the target sum.

10. A low-light image quality adaptive enhancement system for video surveillance, characterized in that, The system includes: The image acquisition module is used to determine the initial monitoring video and obtain the keyframe image corresponding to the initial monitoring video; The image processing module is used to deblur the keyframe image by using the entropy information corresponding to the keyframe image and introducing illumination evaluation parameters, so as to obtain the deblurred image corresponding to the keyframe image. An image adjustment module is used to perform uncertainty analysis on the deblurred image to obtain a target blur value corresponding to the deblurred image, and to perform image adjustment on the deblurred image according to the target blur value to obtain a corresponding uncertain image; An edge detection module is used to perform low-light image enhancement processing on the uncertain image to obtain a target enhanced image; and to perform image edge detection on the target enhanced image to obtain target edge points and the target accuracy corresponding to the target edge points; An enhancement adjustment module is used to perform image enhancement adjustment on the target enhanced image according to the target accuracy until the target enhanced image that meets the preset accuracy is obtained.

Images