A method for visual enhancement of color images

Abstract

The application discloses a color image visual enhancement method, comprising: obtaining an original color image and separating three color components; sharpening each component image by using a Laplace operator, extracting horizontal and vertical gradients by using a Sobel operator, transforming the gradients based on a Gaussian function and completing normalization processing, performing natural logarithm operation on the original component image and the normalized gradient image respectively, adding the obtained two logarithm domain images to perform exponential recovery, restoring to a gray domain, performing Sigmoid function transformation on the recovered component image; calculating the global maximum and minimum values of the Sigmoid enhanced images of the three component images, performing linear stretching, and finally merging the stretched color component images to obtain a color image with visual enhancement. The application can improve the permeability of the color image, improve the brightness of the color image under low illumination, and keep the color effect matching the habit of human eye vision.

Description

Technical Field

[0001] This invention relates to the field of color image processing technology, and in particular to a method for enhancing the visual appearance of color images. Background Technology

[0002] In the field of multi-band imaging, visible light imaging is the imaging method that best matches human visual perception. However, due to poor lighting conditions or low atmospheric transparency, the following problems may occur after image signal processing: (1) The image is too dark, causing the image details to be unsuitable for human visual observation; (2) The transparency is poor, and the image appears as if there is haze.

[0003] Existing technologies, such as patent CN106251300A, describe a fast nighttime foggy image restoration method based on Retinex. First, a guided filter with edge-preserving properties is used to estimate the illuminance image for each color component, and the local contrast coefficient is calculated. Then, Otsu thresholding is performed on the maximum value channel of the original image, and the Gamma parameter of each pixel is adaptively estimated. Gamma correction and contrast preservation are then applied to the illuminance image. Finally, the image is transformed to HSV space, and the contrast of the luminance channel V is enhanced using the CLAHE algorithm. The image is then transformed back to RGB space to obtain the restored image. However, the processing steps, whether guided filtering or the CLAHE algorithm, require significant computational resources, and the enhancement artifacts are obvious, which is incompatible with human visual imaging habits.

[0004] Patent CN110276729A discloses a method for enhancing low-light color images. First, the input image is color-separated to obtain three component images: R, G, and B. Second, homomorphic high-frequency filtering and homomorphic low-frequency filtering are applied to the three component images to obtain high-frequency and low-frequency images for each component, respectively. Next, linear scaling transformation is applied to the low-frequency images of each component. Then, the high-frequency images of each component are processed using a multi-scale Retinex algorithm. Finally, the processed high-frequency and low-frequency images are merged, and the three component images are combined to obtain the enhanced low-light color image. However, the halo effect of the multi-scale Retinex algorithm shows obvious signs of algorithmic enhancement from a human visual perspective, and the sharpness of edge details exhibits a diffusion effect.

[0005] Therefore, a color image visual enhancement method is needed that can achieve image enhancement effects that match human visual habits while ensuring the algorithm is lightweight. Summary of the Invention

[0006] In view of this, the present invention provides a color image visual enhancement method to solve the technical problems of high computational complexity, obvious enhancement traces, diffusion of edge details, and mismatch with human visual habits in the prior art.

[0007] To achieve the above-mentioned technical objectives, the present invention adopts the following technical solution:

[0008] This invention provides a method for visual enhancement of color images, comprising:

[0009] Obtain the original color image, separate the R, G, and B color components from the original color image to obtain three component images;

[0010] The Laplace operator is used to sharpen each component image separately to obtain the sharpened component images.

[0011] Extracting horizontal and vertical gradient maps of the sharpened component images based on the Sobel operator;

[0012] Apply a Gaussian function transformation to the horizontal and vertical gradient maps to generate a Gaussian gradient map.

[0013] Normalize the Gaussian gradient plot to obtain a normalized Gaussian gradient plot.

[0014] Take the natural logarithm of the original component image and the normalized Gaussian gradient image respectively to obtain their respective logarithmic domain images;

[0015] Add the two logarithmic domain images of each component image and then perform an exponential operation to restore it to the grayscale domain, thus obtaining the grayscale domain component image.

[0016] Apply the Sigmoid function transformation to the grayscale component image to obtain the corresponding Sigmoid enhanced image;

[0017] Calculate the maximum and minimum values ​​in the Sigmoid-enhanced image;

[0018] Based on the maximum and minimum values ​​of the Sigmoid enhanced image corresponding to each component image, the global maximum and minimum values ​​of the three component images are obtained, and the component images of each color are linearly stretched according to the global maximum and minimum values.

[0019] The three stretched component images are merged to output a visually enhanced color image.

[0020] Furthermore, the Laplace operator is a 3×3 template, and its convolution kernel is... .

[0021] Furthermore, the Sobel operator includes a horizontal Sobel operator and a vertical Sobel operator;

[0022] The horizontal Sobel operator is ,

[0023] The vertical Sobel operator is .

[0024] Furthermore, the method of Gaussian function transformation is expressed by the following formula:

[0025] ;

[0026] in, This represents a Gaussian gradient plot, where σ represents the Gaussian standard deviation. This is a horizontal gradient plot. This is a vertical gradient plot.

[0027] Furthermore, the method for normalizing the Gaussian gradient map is expressed by the following formula:

[0028] ;

[0029] in, Represents the Gaussian gradient plot. Indicates the image height. Indicates the image width.

[0030] Furthermore, the method of adding the two logarithmic domain images of each component image and then performing an exponential operation to restore it to the grayscale domain, thereby obtaining the grayscale domain component image, can be expressed by the following formula:

[0031]

[0032] in, , representing the original component image Pixel values ​​after natural logarithmic transformation; , representing the original component image The pixel value obtained by multiplying the normalized gradient map point-to-point and then performing a logarithmic transformation; e represents the natural constant.

[0033] Furthermore, the Sigmoid function transformation is applied to the grayscale component image to obtain the corresponding Sigmoid enhanced image, which can be expressed by the formula:

[0034] ;

[0035] in, This represents a grayscale image, and γ is an adjustment parameter used to adjust the enhancement amount.

[0036] Furthermore, the adjustment parameter γ has a value range of 0.1-1.0, with a default value of 0.25.

[0037] Furthermore, the step of obtaining the global maximum and minimum values ​​of the three component images based on the maximum and minimum values ​​of the Sigmoid enhanced image corresponding to each component image, and then linearly stretching the component images of each color according to the global maximum and minimum values, includes:

[0038] Select the set of maximum values ​​of the Sigmoid-enhanced images corresponding to the R, G, and B component images. The maximum value in is the maximum parameter of linear stretching. ;

[0039] Select the set of minimum values ​​of the Sigmoid-enhanced images corresponding to the R, G, and B component images. The minimum value in the value is the minimum parameter of linear stretching. ;

[0040] Based on the maximum value parameter and minimum parameter The component images of each color are linearly stretched, and the method of linear stretching is expressed by the formula:

[0041]

[0042] in, , , Rdst(i,j) represents the Sigmoid-enhanced image corresponding to the R, G, and B component images, respectively. , These represent stretched images after linear stretching of the R, G, and B color channels, respectively.

[0043] Compared with existing technologies, the color image visual enhancement method proposed in this invention has the following advantages:

[0044] (1) This invention uses a comprehensive processing flow of Laplace sharpening, Sobel gradient extraction, Gaussian transformation, log / exponential domain modulation, Sigmoid enhancement and global dynamic range stretching. The Laplace operator enhances details, the Sobel operator extracts structural gradients, Gaussian weights suppress noise, log domain modulation enhances dark information, and the Sigmoid function smooths and enhances brightness differences. Finally, the global maximum and minimum values ​​are combined to uniformly stretch the three RGB channels, so that the image has better brightness levels and color consistency after enhancement. It can take into account both brightness enhancement and contrast enhancement, and can effectively restore image quality under complex conditions such as low light, haze, and backlight.

[0045] (2) For low-contrast color images, this invention highlights the image structure through gradient extraction and Gaussian function transformation, and fuses the original image and local change information in the logarithmic domain, thereby amplifying weak edges and low color difference areas and improving the overall contrast. At the same time, the Sigmoid enhancement step further increases the transition levels between different brightness areas, making the brightness distribution of the image clearer. Combined with the final unified linear stretching, the global brightness range of the image is expanded, improving transparency and color gradation, which can significantly improve the visibility of low-contrast images. It is particularly suitable for scenarios such as video defogging, dusty environment monitoring, and road camera enhancement, enhancing the visual comfort of human observation.

[0046] (3) For low-light color images, this invention enhances pixel differences in dark areas through logarithmic domain transformation, improving details that were previously difficult to discern; after exponential domain recovery, the Sigmoid function is combined to improve overall brightness, brightening dark areas without causing overexposure. Simultaneously, Gaussian gradient weights ensure that noise is not amplified during the enhancement process, helping to maintain image naturalness. By using unified dynamic range stretching across the RGB three channels, the color distribution ratio of dark, bright, and intermediate areas becomes more reasonable, significantly enhancing human vision's ability to recognize low-light scenes and making the outlines, textures, and colors of target objects more distinct. This is particularly suitable for nighttime surveillance, low-light environment shooting, and low-light video enhancement. Attached Figure Description

[0047] Figure 1 A flowchart illustrating a color image visual enhancement method provided by the present invention;

[0048] Figure 2a Low-contrast color image;

[0049] Figure 2b This is a diagram showing the effect of processing a low-contrast color image using this method.

[0050] Figure 3a Low-light color image;

[0051] Figure 3b This is a diagram showing the effect of processing a low-light color image using this method. Detailed Implementation

[0052] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

[0053] Please see Figure 1 This embodiment provides a method for visual enhancement of color images, including:

[0054] Step S101: Obtain the original color image, separate the R, G, and B color components in the original color image to obtain three component images;

[0055] Step S102: Use the Laplace operator to sharpen each component image separately to obtain the sharpened component images;

[0056] Step S103: Extract the horizontal and vertical gradient maps of the sharpened component images based on the Sobel operator;

[0057] Step S104: Perform Gaussian function transformation on the horizontal gradient map and the vertical gradient map to generate a Gaussian gradient map;

[0058] Step S105: Normalize the Gaussian gradient map to obtain a normalized Gaussian gradient map;

[0059] Step S106: Take the natural logarithm of the original component image and the normalized Gaussian gradient map respectively to obtain their respective logarithmic domain images;

[0060] Step S107: Add the two logarithmic domain images of each component image and perform an exponential operation to restore it to the grayscale domain, thus obtaining the grayscale domain component image;

[0061] Step S108: Perform a Sigmoid function transformation on the grayscale component image to obtain the corresponding Sigmoid enhanced image;

[0062] Step S109: Calculate the maximum and minimum values ​​in the Sigmoid-enhanced image;

[0063] Step S110: Based on the maximum and minimum values ​​of the Sigmoid enhanced image corresponding to each component image, obtain the global maximum and minimum values ​​among the three component images, and perform linear stretching on the component images of each color according to the global maximum and minimum values;

[0064] Step S111: Merge the three stretched component images to output a visually enhanced color image.

[0065] The method in this embodiment first uses the LapLace operator for sharpening to ensure that the Sobel operator's gradient calculation can extract details. Second, it performs an exponential transformation by adding the image after logarithmic domain transformation, converting the image to the grayscale domain. Finally, it employs a Sigmoid function transformation and linear contrast stretching to further enhance the image's depth and the differences between different color components, resulting in a more pronounced visual enhancement effect. This method solves both the low contrast problem and the low-light imaging problem of RGB color images. For color images with low contrast, this method can improve contrast and transparency, enhancing the comfort of human visual perception, and is suitable for video dehazing. For low-light color images, this method can increase brightness and contrast between different color components to enhance human visual perception of low-light scenes.

[0066] The following is a detailed description of the processing procedure of the above method using a specific numerical example.

[0067] Assume the original color image as input is Given an image with height nHeight and width nWidth, the processing procedure for this color image is as follows:

[0068] (1) Separate the three color components R, G and B. Each component image is processed separately according to the subsequent steps (2) to (9). The processing flow and parameters are the same. The following uses the red R component image as an example.

[0069] (2) Laplace operator sharpening. The Laplace operator is used to sharpen the image. Perform convolution to obtain the sharpened image. The Laplace convolution operator is as follows:

[0070] .

[0071] Sharpening an image enhances edge details, making the details in the image more apparent and thus improving the clarity of subsequent gradient calculations, enabling the extraction of more useful information.

[0072] (3) Sobel operator extracts gradient map. For the image... The Sobel operator is used for convolution to extract the horizontal gradient RgradHor and the vertical gradient RgradVer; the horizontal and vertical Sobel operators are as follows:

[0073] Horizontal Sobel operator Vertical Sobel operator .

[0074] Using the Sobel operator to calculate the horizontal and vertical gradients of an image can extract edge information, thereby better understanding the structure of the image and helping subsequent processing to enhance image contrast and depth.

[0075] (4) Gaussian function transformation. Using the horizontal and vertical gradient maps, a Gaussian function transformation is performed to obtain the Gaussian gradient map RgradGauss, as shown in the following formula:

[0076]

[0077] in, This represents a Gaussian gradient plot, where σ represents the Gaussian standard deviation. This is a horizontal gradient plot. This is a vertical gradient plot.

[0078] Gaussian transformation can smooth gradient information, remove high-frequency noise in the image, and preserve the integrity of edge information.

[0079] (5) Gaussian gradient map normalization. Calculate the sum of all pixels in the Gaussian gradient map, normalize RgradGauss, and obtain RgradGaussNorm. The formula is as follows:

[0080]

[0081] in, Represents the Gaussian gradient plot. Indicates the image height. Indicates the image width.

[0082] By normalizing the Gaussian gradient map, the gradient information is standardized to avoid values ​​that are too large or too small, while maintaining the stability and naturalness of gradient enhancement, ensuring that the effects of subsequent logarithmic fusion, exponential recovery, and color channel stretching are consistent.

[0083] (6) Take the logarithm Log.

[0084] For the original image and Calculate the natural logarithm to obtain the logarithmic domain image matrix. and The calculation formula is as follows:

[0085]

[0086]

[0087] Since logarithmic transformation can enhance low-brightness areas of an image, converting the image from the linear domain to the logarithmic domain enhances details in low-brightness areas, thereby improving dark details in the image.

[0088] (7) Summation of images and exponentiation. For and The sums are then added together to obtain the summed image, and then an exponential operation is performed to obtain the grayscale image. The calculation is as follows:

[0089]

[0090] in, Represents the original component image Pixel values ​​after natural logarithmic transformation; This represents the pixel value after logarithmic transformation of the gradient map after normalization; e represents the natural constant.

[0091] By performing exponential operations on data in the logarithmic domain, the brightness and contrast of an image can be improved, the dynamic range of the image can be restored, and the image can have richer layers between bright and dark areas.

[0092] (8) Sigmoid function transformation. For The Sigmoid function transformation is calculated using the following formula:

[0093]

[0094] Here, γ is an adjustable parameter responsible for the enhancement amount. It should satisfy (γ>0), where a higher value will result in poorer lighting enhancement but better contrast. Actual verification results show that values ​​between 0.1 and 1.0 have good effects, with the default value of 0.25.

[0095] By adjusting the image contrast using the Sigmoid function, the enhancement effect is made smoother, avoiding over-enhancement, and generating a visual effect that matches human visual habits.

[0096] (9) Calculate the maximum and minimum values. The maximum value MaxRed and the minimum value MinRed of the entire graph.

[0097] (10) Linear stretching. Select the set of maximum values ​​of the three components R, G, and B. The maximum value in is the maximum parameter of linear stretching. Select the set of minimum values ​​for the three components R, G, and B. The minimum value in the value is the minimum parameter of linear stretching. The linear stretching is as follows:

[0098]

[0099] in, , , Rdst(i,j) represents the Sigmoid-enhanced image corresponding to the R, G, and B component images, respectively. , These represent stretched images after linear stretching of the R, G, and B color channels, respectively.

[0100] By linearly stretching and adjusting the maximum and minimum values ​​of each color component, the color gamut of the image is unified, the contrast of the image is optimized, the colors are more vibrant, and the visual effect of the image is enhanced.

[0101] (11) Combine the three color components: R, G, and B.

[0102] Will , , Merged into a three-channel color image This completes the color image visual enhancement processing workflow.

[0103] To verify the actual effectiveness of the present invention, this application processes color images with low contrast and low illumination respectively, and the processing results are as follows. Figure 2a , Figure 2b ,as well as Figure 3a and Figure 3b As shown.

[0104] Specifically, Figure 2a The existing color image suffers from problems such as haziness, blurred details, insufficient color saturation, and an overall grayish appearance. The image processed by the method of this invention is shown below. Figure 2b As shown, Figure 2b It exhibits significantly improved transparency and detail reproduction: fog effects are effectively suppressed, the clarity of distant object outlines is improved, and color gradations are more distinct, meeting the visual expectations of the human eye for natural scenes.

[0105] Figure 3a The color image suffers from insufficient brightness and loss of detail in dark areas. The image enhanced by the method of this invention is shown below. Figure 3b As shown, Figure 3b While improving overall brightness, the algorithm maintains a natural color transition: details in dark areas are effectively restored (such as building textures), and there are no color casts or halo artifacts common in traditional methods, demonstrating the algorithm's ability to balance brightness enhancement and visual comfort.

[0106] As can be seen from the above comparative results, the method of the present invention can effectively solve the problems of low-contrast image fogging and low-light image brightness, and the enhancement result not only preserves the sharpness of edge details, but also conforms to the perceptual characteristics of the human visual system.

[0107] In summary, this embodiment provides a color image visual enhancement method based on multi-stage fusion processing. By separating the RGB components of the original image and then sequentially employing Laplace sharpening, Sobel gradient extraction, Gaussian function weighting, fusion enhancement of the logarithmic and gradient domains, exponential recovery, Sigmoid contrast adjustment, and linear stretching driven by global maximum and minimum values, the method achieves a synergistic enhancement of the original image's brightness, contrast, and detail information. This method effectively enhances the edge structure and local details of the image, improves the overall dynamic range and color levels, maintains a natural visual effect while enhancing brightness, and avoids artifacts and color casts caused by over-enhancement. Through the combined enhancement of multi-source information, it can improve the transparency and clarity of low-contrast images and enhance the brightness and visibility of low-light images, ultimately outputting a high-quality color image that better conforms to human visual habits. The processing method of this invention is applicable to various scenarios such as night monitoring, video defogging, and low-light shooting, and has good practical value and broad application prospects.

[0108] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for visual enhancement of color images, characterized in that, include: Obtain the original color image, separate the R, G, and B color components from the original color image to obtain three component images; The Laplace operator is used to sharpen each component image separately to obtain the sharpened component images. Extracting horizontal and vertical gradient maps of the sharpened component images based on the Sobel operator; Apply a Gaussian function transformation to the horizontal and vertical gradient maps to generate a Gaussian gradient map. Normalize the Gaussian gradient plot to obtain a normalized Gaussian gradient plot. Take the natural logarithm of the original component image and the normalized Gaussian gradient map respectively to obtain their respective logarithmic domain images; Add the two logarithmic domain images of each component image and then perform an exponential operation to restore it to the grayscale domain, thus obtaining the grayscale domain component image. Apply the Sigmoid function transformation to the grayscale component image to obtain the corresponding Sigmoid enhanced image; Calculate the maximum and minimum values ​​in the Sigmoid-enhanced image; Based on the maximum and minimum values ​​of the Sigmoid enhanced image corresponding to each component image, the global maximum and minimum values ​​of the three component images are obtained, and the component images of each color are linearly stretched according to the global maximum and minimum values. The three stretched component images are merged to output a visually enhanced color image.

2. The color image visual enhancement method according to claim 1, characterized in that, The Laplace operator is a 3×3 template, and its convolution kernel is... .

3. The color image visual enhancement method according to claim 1, characterized in that, The Sobel operator includes a horizontal Sobel operator and a vertical Sobel operator; The horizontal Sobel operator is , The vertical Sobel operator is .

4. The color image visual enhancement method according to claim 1, characterized in that, The method of Gaussian function transformation is expressed by the following formula: ； in, This represents a Gaussian gradient plot, where σ represents the Gaussian standard deviation. This is a horizontal gradient plot. This is a vertical gradient plot.

5. The color image visual enhancement method according to claim 4, characterized in that, The method for normalizing the Gaussian gradient map is expressed by the following formula: ； in, Represents the Gaussian gradient plot. Indicates the image height. Indicates the image width.

6. The color image visual enhancement method according to claim 5, characterized in that, The method of adding the two logarithmic domain images of each component image and then performing an exponential operation to restore it to the grayscale domain, thus obtaining the grayscale domain component image, can be expressed by the following formula: in, , representing the original component image Pixel values ​​after natural logarithmic transformation; , representing the original component image The pixel value obtained by multiplying the normalized gradient map point-to-point and then performing a logarithmic transformation; e represents the natural constant.

7. The color image visual enhancement method according to claim 1, characterized in that, The Sigmoid function transformation is applied to the grayscale component image to obtain the corresponding Sigmoid-enhanced image, which can be expressed by the formula: ； in, This represents the grayscale component image, where γ is an adjustment parameter used to adjust the enhancement amount.

8. The color image visual enhancement method according to claim 7, characterized in that, The value range of the adjustment parameter γ is 0.1-1.0, and the default value is 0.

25.

9. The color image visual enhancement method according to claim 1, characterized in that, The process involves obtaining the global maximum and minimum values ​​of the three component images based on the maximum and minimum values ​​of the Sigmoid enhanced image corresponding to each component image, and then linearly stretching the component images of each color according to the global maximum and minimum values, including: Select the set of maximum values ​​of the Sigmoid-enhanced images corresponding to the R, G, and B component images. The maximum value in is the maximum parameter of linear stretching. ; Select the set of minimum values ​​of the Sigmoid-enhanced images corresponding to the R, G, and B component images. The minimum value in the value is the minimum parameter of linear stretching. ; Based on the maximum value parameter and minimum parameter The component images of each color are linearly stretched, and the method of linear stretching is expressed by the formula: in, , , Rdst(i,j) represents the Sigmoid-enhanced image corresponding to the R, G, and B component images, respectively. , These represent stretched images after linear stretching of the R, G, and B color channels, respectively.

Images