Low-illumination image quality improvement method based on adaptive stochastic resonance

Abstract

The invention discloses a low-illumination image quality improvement method based on self-adaptive stochastic resonance. The method is specifically implemented according to the following steps: step 1, synchronously acquiring a grayscale image and a color original image corresponding to the same image through two channels; step 2, preprocessing the grayscale image and the color original image acquired in the step 1; 3, self-adaptive stochastic resonance parameter calculation is carried out; step 4, performing dual-channel adaptive stochastic resonance processing on the preprocessed grayscale image and the preprocessed color original image according to adaptive stochastic resonance parameters to obtain an enhanced image; and 5, carrying out brightness component fusion on the enhanced image, and then carrying out image reconstruction and post-processing. According to the invention, the problem of poor image quality after low-illumination image enhancement in the prior art is solved.

Description

Technical Field

[0001] This invention belongs to the field of image processing methods, and relates to a method for improving the quality of low-light images based on adaptive random resonance. Background Technology

[0002] In low-light scenarios such as security monitoring, nighttime vehicle vision, and mobile photography, it is necessary to enhance the acquired low-light images. Traditional low-light image enhancement techniques have many limitations, as follows: (1) Single-channel imaging defects: Color imaging has insufficient light throughput due to filters, resulting in low signal-to-noise ratio (SNR) and easy noise spots and color distortion; grayscale imaging has a high signal-to-noise ratio, but lacks color information.

[0003] (2) Bottleneck of stochastic resonance technology: The stochastic resonance algorithm with fixed parameters cannot adapt to the differences in local features of the image (such as edges and smooth areas, high / low signal-to-noise ratio areas), and it is easy to over-amplify noise while enhancing the signal.

[0004] Therefore, there is an urgent need for a low-light image enhancement scheme that can dynamically adjust parameters based on image features and deeply integrate dual-channel information. Summary of the Invention

[0005] The purpose of this invention is to provide a method for improving the quality of low-light images based on adaptive stochastic resonance, which solves the problem of poor image quality after low-light image enhancement caused by the bottlenecks of single-channel imaging and stochastic resonance technology in the prior art.

[0006] The technical solution adopted in this invention is a low-light image quality improvement method based on adaptive stochastic resonance, which is implemented according to the following steps: Step 1: Simultaneously acquire grayscale images corresponding to the same image using dual channels. and original color image ; Step 2: Preprocess the grayscale image and the original color image acquired in Step 1; Step 3: Perform adaptive stochastic resonance parameter calculation; Step 4: Perform dual-channel adaptive stochastic resonance processing on the preprocessed grayscale image and the original color image based on the adaptive stochastic resonance parameters to obtain the enhanced image; Step 5: Perform luminance component fusion on the enhanced image, and then perform image reconstruction and post-processing.

[0007] Preferably, the preprocessing of the acquired grayscale image in step 2 specifically includes: The acquired grayscale image is normalized: Will Pixel values ​​mapped to The interval is used to obtain the normalized grayscale image. Specifically:

[0008] in, yes minimum pixel value and yes Maximum pixel value, and By traversal Full image pixel acquisition, grayscale image Centered at pixel Pixel value at that location, The grayscale image after normalization is located at pixels. Pixel value at; Step 2 involves the acquisition of the original color image. The preprocessing process specifically involves: Step 2.1, process the original color image To perform de-mosaic processing, specifically: Bilinear interpolation is used to restore the image to RGB. Step 2.2 involves converting the RGB image to a different color space, specifically as follows: Convert the RGB image to the YUV color space and separate the luminance component. Blue chromaticity component and red chromaticity component :

[0009] Wherein, R, G, and B are the R channel values, G channel values, and B channel values ​​of the RGB image, respectively.

[0010] Preferably, step 3 specifically includes: Step 3.1: Extract the normalized grayscale images respectively. With brightness component Local features; Step 3.2: Calculate the adaptive stochastic resonance parameters based on the extracted local features.

[0011] Preferably, step 3.1 specifically includes: Defined by pixels Centered sliding window Calculate the grayscale image within the sliding window respectively. With brightness component Luminance variance Texture complexity Local signal-to-noise ratio and grayscale images Structural salience ; Among them, the luminance variance The formula is as follows:

[0012] in, or , For the corresponding sliding window Any pixel within, grayscale image In the sliding window any pixel within pixel value, Luminance component In the sliding window any pixel within Pixel brightness value at that location, This represents the average brightness within the corresponding window.

[0013] Where N is the sliding window The total number of pixels within; Texture complexity The formula is as follows:

[0014] in,

[0015] in, , Sliding window for the corresponding image Inner pixel The directional gradient in the x and y directions, when calculating a grayscale image. Texture complexity hour, , Grayscale images sliding window Inner pixel The directional gradient in the x and y directions; when calculating the brightness component. Texture complexity hour, , These are the luminance components. sliding window Inner pixel The directional gradient in the x and y directions; Local signal-to-noise ratio The formula is as follows:

[0016] in, It is the minimum value; The unit is dB; Structural salience The formula is as follows:

[0017] in, for Structure mask image At pixel Pixel value at the location, for Edge intensity map At pixel Edge strength value at the location; Structure mask image The edge intensity map E is obtained as follows: Step a, for Binarized edge images are generated using Canny edge detection. ; Step b, for Perform morphological dilation to generate a structure mask image. ; Step c: Calculate the structure mask image using the Sobel operator. Then calculate the edge intensity map based on the gradients in the x and y directions. :

[0018] in, , These are structure mask images. Gradient in the x and y directions, For edge intensity map E at pixel point The edge strength value at that location.

[0019] Preferably, step 3.2 specifically includes: System stability parameters are obtained based on structural saliency and luminance variance. The formula is as follows:

[0020] in, As the baseline parameter, The structural significance adjustment coefficient. This is the brightness variance adjustment coefficient; grayscale image With brightness component Luminance variance Substitute respectively The formulas are used to obtain grayscale images respectively. With brightness component Corresponding system stability parameters ; Based on local signal-to-noise ratio Obtain the nonlinear strength parameters of the system :

[0021] in, As the baseline parameter, This is the SNR adjustment coefficient. The maximum signal-to-noise ratio threshold; grayscale image With brightness component Local signal-to-noise ratio Substitute respectively The formulas are used to obtain grayscale images respectively. With brightness component Corresponding nonlinear intensity parameters .

[0022] Preferably, in step 4, adaptive random resonance processing is performed on the preprocessed grayscale image based on the adaptive random resonance parameters, specifically as follows: (1) Construct an adaptive bistable stochastic resonance model, specifically as follows:

[0023] in, grayscale image after normalization , as the input signal of the adaptive bistable stochastic resonance model; It is Gaussian white noise that follows a normal distribution. , For time step; and These correspond to the grayscale images obtained in step 3.2. The corresponding parameters; grayscale image noise intensity :

[0024] grayscale image corresponding and The grayscale image obtained by substituting into the above formula noise intensity ; in, As the reference noise intensity, , This is the SNR adjustment coefficient. This is a texture complexity adjustment factor. ; (2) The adaptive bistable stochastic resonance model is solved iteratively using the improved Euler method formula. The iterative formula is as follows:

[0025] in, , is the right-hand side term of the differential equation; The iteration termination condition is that the condition is met for three consecutive iterations. ; (3) After the iteration converges Mapping back to the original pixel value range yields the enhanced grayscale image. The formula is as follows:

[0026] in, To enhance the grayscale image any pixel The pixel value.

[0027] Preferably, step 4, which involves adaptive random resonance processing of the preprocessed original color image based on adaptive random resonance parameters, specifically includes: For the brightness channel Adaptive enhancement is performed, specifically as follows: (1) Construct an adaptive bistable stochastic resonance model, specifically as follows:

[0028] in, For brightness channel , as the input signal of the adaptive bistable stochastic resonance model; It is Gaussian white noise that follows a normal distribution. ; and These correspond to the brightness channels obtained in step 3.2. The corresponding parameters; in, Luminance component The final noise intensity is obtained as follows: Brightness component of and The following formula yields the luminance component. noise intensity :

[0029] in, As the reference noise intensity, , This is the SNR adjustment coefficient. This is a texture complexity adjustment factor. ; Then, the luminance component... noise intensity By introducing a color influence coefficient to correct the noise intensity, the luminance component is obtained. final noise intensity :

[0030] in, , , Luminance component At pixel The pixel brightness value at that location; (2) Then, the adaptive bistable stochastic resonance model is solved iteratively using the improved Euler method formula. The iterative formula is as follows:

[0031] However, , is the right-hand side term of the differential equation; The iteration termination condition is that the condition is met for three consecutive iterations. ; (3) After the iteration converges Mapping back to the original pixel value range yields the enhanced luminance channel. The formula is as follows:

[0032] in, To enhance the brightness channel any pixel pixel brightness value, Luminance component Maximum brightness value, Luminance component The minimum brightness value; (4) For the chroma channel , Adaptive noise reduction is used to obtain the corresponding enhanced chroma channel, specifically: Constructing a multi-steady-state stochastic resonance model:

[0033] in, For the input chromaticity components, The signal amplification factor is 1. Based on the threshold, For adaptive adjustment coefficient, The chromaticity auxiliary noise follows a normal distribution. ; The calculation yields Mapping back to the original chroma range yields the enhanced chroma channel. and Specifically: for Channel: Mapping formula is ; for Channel: Mapping formula is ; in, For chroma channels any pixel chromaticity value, For chroma channels At any pixel point The chromaticity value.

[0034] Preferably, step 5, which involves fusion of the luminance components in the enhanced image, specifically includes: Weighted fusion and The fused luminance components are obtained. :

[0035] in, for any pixel pixel values, for any pixel pixel brightness value, Luminance component any pixel Pixel brightness value; ; ; in, .

[0036] Preferably, the image reconstruction and post-processing in step 5 specifically includes: The fused luminance components Enhance color channels , Convert back to RGB space to obtain a preliminary fused image. The conversion formula is as follows:

[0037] Then the preliminary fused images Perform adaptive gamma correction, gamma value Based on global average brightness Adjustment: hour ; hour ; hour ; Then, local histogram equalization is performed on the image after adaptive gamma correction to output the final enhanced image. .

[0038] The beneficial effects of this invention are: This invention uses a single-aperture beam splitting design to simultaneously acquire high signal-to-noise ratio grayscale and color images. The grayscale channel provides clear details, while the color channel can fully preserve color information. Combined with subsequent fusion algorithms, it achieves dual protection of "detail presentation" and "color preservation", thus solving the defects of single-channel imaging.

[0039] This invention abandons the traditional fixed parameter mode and dynamically adjusts the core parameters of random resonance based on local image features. It enhances the signal enhancement effect in areas with weak signals and suppresses noise amplification in areas with high signal quality, achieving a dynamic balance between "detail enhancement" and "noise suppression". This avoids the effect deviation caused by adapting a single parameter to all areas and breaks through the bottleneck of random resonance technology.

[0040] In summary, this invention achieves synergistic optimization of low-light image detail enhancement, noise suppression, and color fidelity by dynamically adjusting random resonance parameters and deeply fusing dual-channel image information, thereby improving the overall image quality while meeting real-time and engineering practicality requirements. Attached Figure Description

[0041] Figure 1 This is a flowchart of the low-light image quality improvement method based on adaptive stochastic resonance of the present invention; Figure 2 This is a schematic diagram of the hardware architecture of a specific implementation of the low-light image quality improvement method based on adaptive stochastic resonance of the present invention. Detailed Implementation

[0042] The following detailed description is provided in conjunction with specific implementation methods.

[0043] Example 1 This invention relates to a low-light image quality enhancement method based on adaptive stochastic resonance, the process of which is as follows: Figure 1 As shown, the specific steps are as follows: Step 1: Simultaneously acquire grayscale images corresponding to the same image using dual channels. and original color image ; Step 2: Preprocess the grayscale image and the original color image acquired in Step 1; Step 3: Perform adaptive stochastic resonance parameter calculation; Step 4: Perform dual-channel adaptive stochastic resonance processing on the preprocessed grayscale image and the original color image based on the adaptive stochastic resonance parameters to obtain the enhanced image; Step 5: Perform luminance component fusion on the enhanced image, and then perform image reconstruction and post-processing.

[0044] Example 2 Based on Example 1, the preprocessing of the acquired grayscale image in step 2 specifically involves: The acquired grayscale image is normalized: Will Pixel values ​​mapped to The range is adjusted to eliminate the influence of pixel value magnitude differences on subsequent stochastic resonance calculations, resulting in a normalized grayscale image. Specifically:

[0045] in, yes minimum pixel value and yes Maximum pixel value, and By traversal Full image pixel acquisition, grayscale image Centered at pixel Pixel value at that location, The grayscale image after normalization is located at pixels. Pixel value at; Step 2 involves the acquisition of the original color image. The preprocessing process specifically involves: Step 2.1, process the original color image To perform de-mosaic processing, specifically: Bilinear interpolation is used to restore the image to RGB. Original color image It is in Bayer format, where each pixel contains only a single RGB component, and needs to be restored to a full-resolution RGB image through bilinear interpolation. For R channel pixels (coordinates) (For odd-numbered rows and odd-numbered columns): its missing G component is calculated by weighted average of the four adjacent G pixels (top, bottom, left, and right), and its B component is calculated by weighted average of the four adjacent B pixels (diagonally). For G channel pixels (coordinates) (For odd-numbered rows and even-numbered columns or even-numbered rows and odd-numbered columns): its missing R component is calculated by averaging the two adjacent R pixels on the left and right, and the B component is calculated by averaging the two adjacent B pixels on the top and bottom. For B channel pixels (coordinates) (For even-numbered rows and even-numbered columns): its missing R component is calculated by weighted average of the four adjacent R pixels diagonally, and its G component is calculated by weighted average of the four adjacent G pixels above, below, left, and right. Through the above interpolation, a full-resolution RGB image is output (R, G, and B channels each have a depth of 12 bits, and the pixel value range is 0-4095). Step 2.2 involves converting the RGB image to a different color space, specifically as follows: Convert the RGB image to the YUV color space and separate the luminance component. Blue chromaticity component and red chromaticity component This enables independent processing of "brightness enhancement" and "color fidelity preservation":

[0046] Where R, G, and B are the R channel values, G channel values, and B channel values ​​of the RGB image, respectively. The original luminance component (value range 0-1) reflects the brightness and darkness information of the image and is the core object of enhancement; This represents the blue chromaticity component (value range: -0.436 to 0.436). The red chromaticity component (range -0.615 to 0.615) and the red chromaticity component together determine the image color, and stability must be maintained while reducing noise.

[0047] Example 3 Based on Example 2, step 3 specifically includes: Based on a 32×32 sliding window (window step size 16 pixels to avoid redundant calculations), local features are extracted from the grayscale image and the color luminance channel respectively, and then the core parameters of random resonance (noise intensity) are calculated. Bistable system parameters and This achieves dynamic mapping of "local features - parameters", specifically: Step 3.1: Extract the normalized grayscale images respectively. With brightness component Local features; specifically: Defined by pixels The 32x32 area centered on it is the sliding window. The total number of pixels in the window Calculate the grayscale image within the sliding window respectively. With brightness component Luminance variance Texture complexity Local signal-to-noise ratio and grayscale images Structural salience ; Among them, the luminance variance This reflects the degree of brightness fluctuation within the window. Greater fluctuations indicate potentially higher noise levels in the area. The formula is as follows:

[0048] in, or , For the corresponding sliding window Any pixel within, grayscale image In the sliding window any pixel within pixel value, Luminance component In the sliding window any pixel within Pixel brightness value at that location, This represents the average brightness within the corresponding window.

[0049] Where N is the sliding window The total number of pixels within; Texture complexity The gradient magnitude reflects the density of texture within the window; the more complex the texture (such as vegetation or text), the larger the value. The formula is as follows:

[0050] in,

[0051] in, , Sliding window for the corresponding image Inner pixel The directional gradient in the x and y directions, when calculating a grayscale image. Texture complexity hour, , Grayscale images sliding window Inner pixel The directional gradient in the x and y directions; when calculating the brightness component. Texture complexity hour, , These are the luminance components. sliding window Inner pixel The directional gradient in the x and y directions; Local signal-to-noise ratio The ratio of signal to noise within the evaluation window is used; a higher value indicates better signal quality. The formula is as follows:

[0052] in, To minimize, avoid When the denominator is 0; The unit is dB (areas exceeding this value are considered high signal-to-noise ratio areas and do not require signal enhancement). Structural salience By combining the structure mask of the grayscale channel with the edge strength, the saliency of the structure (edges, textures) within the window is quantized using the following formula:

[0053] in," "This is pixel-level multiplication; for Structure mask image At pixel Pixel value at the location, for Edge intensity map At pixel Edge strength value at the location, The value range is [0,√2], and the larger the value, the more significant the region structure (such as object edges and detailed textures). Structure mask image The edge intensity map E is obtained as follows: Step a, for Binarized edge images are generated using Canny edge detection. Specifically: (1) A 5×5 Gaussian filter (standard deviation σ=1.4) was used for... Smoothing is performed to suppress noise interference with edge detection; (2) Calculate the gradient magnitude of the smoothed image (using the Sobel operator, x-direction operator) y-direction operator ), to obtain the maximum gradient. ; (3) Set dual thresholds: low threshold High threshold The gradient magnitude is greater than The pixels are marked as "strong edges", which are between and Pixels connected to strong edges are labeled as "weak edges," and the rest are background, generating a binarized edge image. (Pixel value 0 represents the background, 1 represents the edge); Step b, for Perform morphological dilation to generate a structure mask image. Specifically: Using 3×3 rectangular structural elements ( )right Perform morphological dilation to fill edge gaps (such as thin edges and edge breaks) and generate a structural mask. (Pixel value 0 represents an unstructured area, and 1 represents a structured area), ensuring the integrity of the edge area; Step c: Calculate the structure mask image using the Sobel operator. Then calculate the edge intensity map based on the gradients in the x and y directions. :

[0054] in, , These are structure mask images. Gradient in the x and y directions, For edge intensity map E at pixel point Edge strength value at the location, The value range is [0,√2], and the larger the value, the sharper the edge; Step 3.2: Calculate the adaptive stochastic resonance parameters based on the extracted local features, specifically as follows: System stability parameters are obtained based on structural saliency and luminance variance. Significant structural areas need to be improved. To enhance edge sharpness, high-brightness variance areas need to be reduced. To suppress noise, The formula is as follows:

[0055] in, As the baseline parameter, The structural significance adjustment coefficient. This is the brightness variance adjustment coefficient; grayscale image With brightness component Luminance variance Substitute respectively The formulas are used to obtain grayscale images respectively. With brightness component Corresponding system stability parameters ; Based on local signal-to-noise ratio Obtain the nonlinear strength parameters of the system Low SNR areas need to be improved To enhance nonlinear signal amplification, the SNR needs to be reduced in the high SNR region. To avoid over-enhancement, The formula is as follows:

[0056] in, As the baseline parameter, This is the SNR adjustment coefficient. The maximum signal-to-noise ratio threshold; grayscale image With brightness component Local signal-to-noise ratio Substitute respectively The formulas are used to obtain grayscale images respectively. With brightness component Corresponding nonlinear intensity parameters .

[0057] Example 4 Based on Example 3, step 4 involves performing adaptive random resonance processing on the preprocessed grayscale image according to the adaptive random resonance parameters, specifically as follows: (1) Construct an adaptive bistable stochastic resonance model, specifically as follows:

[0058] in, grayscale image after normalization , as the input signal of the adaptive bistable stochastic resonance model; It is Gaussian white noise that follows a normal distribution. , For time step; and These correspond to the grayscale images obtained in step 3.2. The corresponding parameters; grayscale image noise intensity To control the intensity of auxiliary noise in the stochastic resonance model, it needs to be dynamically adjusted based on the local signal-to-noise ratio and texture complexity: low SNR regions require increased... To enhance the signal, high-texture areas need to be reduced. To avoid noise amplification, The formula is:

[0059] grayscale image corresponding and The grayscale image obtained by substituting into the above formula noise intensity ; in, As the reference noise intensity, (Determined through optimization testing of 100 sets of low-illuminance samples (illuminance 0.1-10 lux)) This is the SNR adjustment coefficient (a negative sign indicates a higher SNR). smaller), This is a texture complexity adjustment factor. (A plus sign indicates a more complex texture,) Slightly higher to retain details); (2) The adaptive bistable stochastic resonance model is solved iteratively using the improved Euler method formula, balancing computational accuracy and real-time performance. The iterative formula is as follows:

[0060] in, , is the right-hand side term of the differential equation; The iteration termination condition is that the condition is met for three consecutive iterations. (To ensure convergence to a stable value), the number of iterations is usually 5-10 to avoid overcomputation; (3) After the iteration converges (Value range [-1, 1]) Mapping back to the original pixel value range yields the enhanced grayscale image. The formula is as follows:

[0061] in, To enhance the grayscale image any pixel The pixel value.

[0062] Step 4 involves performing adaptive stochastic resonance processing on the preprocessed color original image based on the adaptive stochastic resonance parameters. Specifically, this process is as follows: Color image processing is divided into luminance channel enhancement and chrominance channel noise reduction, which respectively adopt bistable and multistable models to ensure that brightness is improved while avoiding color distortion. Among them, for the luminance channel Adaptive enhancement is performed, specifically as follows: (1) Construct an adaptive bistable stochastic resonance model, specifically as follows:

[0063] in, For brightness channel , as the input signal of the adaptive bistable stochastic resonance model; It is Gaussian white noise that follows a normal distribution. ; and These correspond to the brightness channels obtained in step 3.2. The corresponding parameters; in, Luminance component The final noise intensity is obtained as follows: Brightness component of and The following formula yields the luminance component. noise intensity :

[0064] in, As the reference noise intensity, , This is the SNR adjustment coefficient. This is a texture complexity adjustment factor. ; Then, the luminance component... noise intensity By introducing a color influence coefficient to correct the noise intensity, the luminance component is obtained. final noise intensity :

[0065] in, , , Luminance component At pixel The pixel brightness value at that location is corrected to avoid loss of detail caused by excessive noise reduction in high-brightness areas; (2) Then, the adaptive bistable stochastic resonance model is solved iteratively using the improved Euler method formula. The iterative formula is as follows:

[0066] However, , is the right-hand side term of the differential equation; The iteration termination condition is that the condition is met for three consecutive iterations. ; (3) After the iteration converges Mapping back to the original pixel value range yields the enhanced luminance channel. The formula is as follows:

[0067] in, To enhance the brightness channel any pixel pixel brightness value, Luminance component Maximum brightness value, Luminance component The minimum brightness value; Among them, for the chroma channel , Adaptive noise reduction is used to obtain the corresponding enhanced chroma channel, specifically: By leveraging the "soft saturation" property of the tanh function, noise is compressed to a reasonable range while preserving the chromaticity signal, thus constructing a multistable stochastic resonance model:

[0068] in, For the input chromaticity components, This is a signal amplification factor, used to enhance weak chroma signals; Use a base threshold to avoid function value overflow caused by an excessively small denominator; This is an adaptive adjustment coefficient that varies with the absolute value of chromaticity. Dynamic adjustment of the denominator: In high chromaticity regions (such as vivid colors), the denominator is increased to reduce signal compression; in low chromaticity regions (such as pale colors), the denominator is decreased to enhance noise suppression. The chromaticity auxiliary noise follows a normal distribution. ; The calculation yields Mapping back to the original chroma range yields the enhanced chroma channel. and Specifically: for Channel: Mapping formula is ; for Channel: Mapping formula is ; in, For chroma channels any pixel chromaticity value, For chroma channels At any pixel point The chromaticity value.

[0069] Example 5 Based on Example 4, step 5 involves fusion of the luminance components in the enhanced image, specifically as follows: Weighted fusion and The fused luminance components are obtained. :

[0070] in, for any pixel pixel values, for any pixel pixel brightness value, Luminance component any pixel Pixel brightness value; ;

[0071] in, ; Image reconstruction and post-processing specifically involve: The fused luminance components Enhance color channels , Convert back to RGB space to obtain a preliminary fused image. The conversion formula is as follows:

[0072] The converted RGB components range from 0 to 1 and need to be mapped to 0-255 (8-bit depth). The formula is as follows: The same applies to the G and B channels.

[0073] Then the preliminary fused images Adaptive gamma correction is performed, specifically as follows: Based on the global average brightness of the image Adjusting the gamma value To solve the problem of overall darkness under low light conditions, the steps are as follows: Calculate the global average brightness: ; Dynamic adjustment : like (Extremely dark environment) Enhance the brightness of dark areas; like (Low-light environment) Balance the light and dark areas; like (Low-light environment) Avoid overexposure of bright areas; Gamma correction formula: The same applies to the G and B channels.

[0074] Next, local histogram equalization is performed on the image after adaptive gamma correction. Specifically, a 16×16 sliding window is used for contrast-limited histogram equalization to further enhance local details. The parameter settings are as follows: A. Window step size of 8 pixels to ensure full image coverage; B. Limit the contrast threshold to 2.0 to avoid artifacts caused by excessive local enhancement; Output the final enhanced image .

[0075] Example 6 Based on Example 5, This invention employs a single-aperture beam splitting imaging system for image acquisition, such as... Figure 2 As shown, a single optical aperture lens (50mm focal length, F / 1.4 aperture) is used, along with a 50:50 split-beam mirror (1mm thickness, 95% transmittance). A high-precision beam splitter with a 50:50 ± 5% split ratio is then applied to split the light into two paths. The reflected light path is transmitted to a grayscale sensor (without a filter, receiving full-spectrum light from 400-900nm) to acquire a high signal-to-noise ratio grayscale image. Transmission optical path: transmitted to a color sensor (equipped with a Bayer filter, response band 400-700nm) to acquire raw color images. The FPGA generates a TTL level trigger signal (frequency 50Hz±1Hz) through a synchronization circuit to control the time synchronization error of the two sensors (grayscale sensor and color sensor) to be ≤1ms and the spatial alignment deviation to be ≤1 pixel. The grayscale sensor uses a back-illuminated CMOS (2.4μm pixel size, 1200×1920 resolution); the color sensor uses a Bayer filter CMOS (2.8μm pixel size, 1200×1920 resolution); the synchronization circuit uses an FPGA model XC7Z020 with a trigger signal frequency of 50Hz to ensure that the acquisition time difference between the two sensors is <10μs.

[0076] During data acquisition, the grayscale sensor and color sensor are exposed synchronously, and grayscale images are output synchronously. With the original color image .

Claims

1. A method for improving the quality of low-light images based on adaptive stochastic resonance, characterized in that, The specific steps are as follows: Step 1: Simultaneously acquire grayscale images corresponding to the same image using dual channels. and original color image ; Step 2: Preprocess the grayscale image and the original color image acquired in Step 1; Step 3: Perform adaptive stochastic resonance parameter calculation; Step 4: Perform dual-channel adaptive stochastic resonance processing on the preprocessed grayscale image and the original color image based on the adaptive stochastic resonance parameters to obtain the enhanced image; Step 5: Perform luminance component fusion on the enhanced image, and then perform image reconstruction and post-processing.

2. The low-light image quality improvement method based on adaptive stochastic resonance according to claim 1, characterized in that, The preprocessing of the acquired grayscale image in step 2 specifically involves: The acquired grayscale image is normalized: Will Pixel values ​​mapped to The interval is used to obtain the normalized grayscale image. Specifically: in, yes minimum pixel value and yes Maximum pixel value, and By traversal Full image pixel acquisition, grayscale image Centered at pixel Pixel value at that location, The grayscale image after normalization is located at pixels. Pixel value at; In step 2, the acquired original color image The preprocessing process specifically involves: Step 2.1, process the original color image To perform de-mosaic processing, specifically: Bilinear interpolation is used to restore the image to RGB. Step 2.2 involves converting the RGB image to a different color space, specifically as follows: Convert the RGB image to the YUV color space and separate the luminance component. Blue chromaticity component and red chromaticity component : Wherein, R, G, and B are the R channel values, G channel values, and B channel values ​​of the RGB image, respectively.

3. The low-light image quality improvement method based on adaptive stochastic resonance according to claim 2, characterized in that, Step 3 specifically involves: Step 3.1: Extract the normalized grayscale images respectively. With brightness component Local features; Step 3.2: Calculate the adaptive stochastic resonance parameters based on the extracted local features.

4. The low-light image quality improvement method based on adaptive stochastic resonance according to claim 3, characterized in that, Step 3.1 specifically involves: Defined by pixels Centered sliding window Calculate the grayscale image within the sliding window respectively. With brightness component Luminance variance Texture complexity Local signal-to-noise ratio and grayscale images Structural salience ; Among them, the luminance variance The formula is as follows: in, or , For the corresponding sliding window Any pixel within, grayscale image In the sliding window any pixel within pixel value, Luminance component In the sliding window any pixel within Pixel brightness value at that location, This represents the average brightness within the corresponding window. Where N is the sliding window The total number of pixels within; The texture complexity The formula is as follows: in, in, , Sliding window for the corresponding image Inner pixel The directional gradient in the x and y directions, when calculating a grayscale image. Texture complexity hour, , Grayscale images sliding window Inner pixel The directional gradient in the x and y directions; when calculating the brightness component. Texture complexity hour, , These are the luminance components. sliding window Inner pixel The directional gradient in the x and y directions; The local signal-to-noise ratio The formula is as follows: in, It is the minimum value; The unit is dB; The structural salience The formula is as follows: in, for Structure mask image At pixel Pixel value at the location, for Edge intensity map At pixel Edge strength value at the location; The structural mask image The edge intensity map E is obtained as follows: Step a, for Binarized edge images are generated using Canny edge detection. ; Step b, for Perform morphological dilation to generate a structure mask image. ; Step c: Calculate the structure mask image using the Sobel operator. Then calculate the edge intensity map based on the gradients in the x and y directions. : in, , These are structure mask images. Gradient in the x and y directions, For edge intensity map E at pixel point The edge strength value at that location.

5. The low-light image quality improvement method based on adaptive stochastic resonance according to claim 4, characterized in that, Step 3.2 specifically involves: System stability parameters are obtained based on structural saliency and luminance variance. The formula is as follows: in, As the baseline parameter, The structural significance adjustment coefficient. This is the brightness variance adjustment coefficient; grayscale image With brightness component Luminance variance Substitute respectively The formulas are used to obtain grayscale images respectively. With brightness component Corresponding system stability parameters ; Based on local signal-to-noise ratio Obtain the nonlinear strength parameters of the system : in, As the baseline parameter, This is the SNR adjustment coefficient. The maximum signal-to-noise ratio threshold; grayscale image With brightness component Local signal-to-noise ratio Substitute respectively The formulas are used to obtain grayscale images respectively. With brightness component Corresponding nonlinear intensity parameters .

6. The low-light image quality improvement method based on adaptive stochastic resonance according to claim 5, characterized in that, In step 4, adaptive random resonance processing is performed on the preprocessed grayscale image based on the adaptive random resonance parameters, specifically as follows: (1) Construct an adaptive bistable stochastic resonance model, specifically as follows: in, grayscale image after normalization , as the input signal of the adaptive bistable stochastic resonance model; It is Gaussian white noise that follows a normal distribution. , For time step; and These correspond to the grayscale images obtained in step 3.

2. The corresponding parameters; grayscale image noise intensity : grayscale image corresponding and The grayscale image obtained by substituting into the above formula noise intensity ; in, As the reference noise intensity, , This is the SNR adjustment coefficient. This is a texture complexity adjustment factor. ; (2) The adaptive bistable stochastic resonance model is solved iteratively using the improved Euler method formula. The iterative formula is as follows: in, , is the right-hand side term of the differential equation; The iteration termination condition is that the condition is met for three consecutive iterations. ; (3) After the iteration converges Mapping back to the original pixel value range yields the enhanced grayscale image. The formula is as follows: in, To enhance the grayscale image any pixel The pixel value.

7. The low-light image quality improvement method based on adaptive stochastic resonance according to claim 6, characterized in that, In step 4, the adaptive random resonance processing of the preprocessed color original image based on the adaptive random resonance parameters is specifically as follows: For the brightness channel Adaptive enhancement is performed, specifically as follows: (1) Construct an adaptive bistable stochastic resonance model, specifically as follows: in, For brightness channel , as the input signal of the adaptive bistable stochastic resonance model; It is Gaussian white noise that follows a normal distribution. ; and These correspond to the brightness channels obtained in step 3.

2. The corresponding parameters; in, Luminance component The final noise intensity is obtained as follows: Brightness component of and The following formula yields the luminance component. noise intensity : in, As the reference noise intensity, , This is the SNR adjustment coefficient. This is a texture complexity adjustment factor. ; Then, the luminance component... noise intensity By introducing a color influence coefficient to correct the noise intensity, the luminance component is obtained. final noise intensity : in, , , Luminance component At pixel The pixel brightness value at that location; (2) Then, the adaptive bistable stochastic resonance model is solved iteratively using the improved Euler method formula. The iterative formula is as follows: However, , is the right-hand side term of the differential equation; The iteration termination condition is that the condition is met for three consecutive iterations. ; (3) After the iteration converges Mapping back to the original pixel value range yields the enhanced luminance channel. The formula is as follows: in, To enhance the brightness channel any pixel pixel brightness value, Luminance component Maximum brightness value, Luminance component The minimum brightness value; (4) For the chroma channel , Adaptive noise reduction is used to obtain the corresponding enhanced chroma channel, specifically: Constructing a multi-steady-state stochastic resonance model: in, For the input chromaticity components, The signal amplification factor is 1. Based on the threshold, For adaptive adjustment coefficient, The chromaticity auxiliary noise follows a normal distribution. ; The calculation yields Mapping back to the original chroma range yields the enhanced chroma channel. and Specifically: for Channel: Mapping formula is ; for Channel: Mapping formula is ; in, For chroma channels any pixel chromaticity value, For chroma channels At any pixel point The chromaticity value.

8. The low-light image quality enhancement method based on adaptive stochastic resonance according to claim 7, wherein step 5, fusion of the luminance component in the enhanced image, specifically comprises: Weighted fusion and The fused luminance components are obtained. : in, for any pixel pixel values, for any pixel pixel brightness value, Luminance component any pixel Pixel brightness value; ; ； in, .

9. The low-light image quality enhancement method based on adaptive stochastic resonance according to claim 8, wherein the image reconstruction and post-processing in step 5 specifically comprises: The fused luminance components Enhance color channels , Convert back to RGB space to obtain a preliminary fused image. The conversion formula is as follows: Then the preliminary fused images Perform adaptive gamma correction, gamma value Based on global average brightness Adjustment: hour ; hour ; hour ; Then, local histogram equalization is performed on the image after adaptive gamma correction to output the final enhanced image. .

Images