CN114429426B - Low-illumination image quality improvement method based on Retinex model - Google Patents

Abstract

The invention relates to a low-light image quality improvement method based on a Retinex model, which belongs to the technical field of image processing. The invention solves the problem that the quality of the obtained image is poor when using the existing low-illumination image quality improvement algorithm to process the low-illumination image. The invention firstly obtains the detail layer image and the illumination layer image by layering the digital image through the Retinex model; secondly, a nonlinear global brightness mapping function is designed to map the illumination layer image to obtain the illumination layer enhanced image; The detail layer image mapping function is used to stretch the detail layer image to obtain the detail layer enhanced image; finally, each pixel of the detail layer enhanced image and the illumination layer enhanced image is multiplied to synthesize the low illumination enhanced image. The method of the present invention can be applied to improve the quality of low-light images.

Description

A low-light image quality improvement method based on Retinex model

technical field

The invention belongs to the technical field of image processing, and in particular relates to a low-illumination image quality improvement method based on a Retinex model.

Background technique

Due to the fact that light is easily affected by the environment in the propagation path, the images captured by digital cameras have uneven brightness distribution, unclear details and textures in low-illumination areas, and poor image quality and visual effects for the human eye. Therefore, how to improve the influence of low illumination effect on image quality has become a hot issue in the field of image processing in recent years.

The classic low-illumination image quality improvement algorithms mainly include image grayscale segmentation mapping algorithm, histogram equalization algorithm, and Gamma correction algorithm. Although the classic low-light image quality improvement algorithm can suppress the visual problems caused by the low-light effect to a certain extent, the selection of free parameters, excessive image enhancement, and relatively blurred details affect the image quality of the enhanced image. Therefore, the use of After the existing low-illumination image quality improvement algorithms process the low-illumination image, the quality of the obtained image is still poor.

SUMMARY OF THE INVENTION

The purpose of the present invention is to solve the problem of poor image quality when using the existing low-illumination image quality improvement algorithm to process the low-illumination image, and propose a low-illumination image quality improvement method based on the Retinex model.

The technical scheme that the present invention takes to solve the above-mentioned technical problems is:

A low-light image quality improvement method based on a Retinex model, the method specifically comprises the following steps:

Step 1. After converting the acquired image from the RGB channel to the HSV channel, the data of the V channel, the H channel and the S channel are obtained respectively;

Then decompose the V channel data into the light layer image and the detail layer image;

Step 2: Perform global brightness mapping processing on the illumination layer image to obtain an illumination layer image after global brightness mapping;

Then perform pixel value domain expansion on the illumination layer image after global brightness mapping, and obtain the illumination layer image after pixel value domain expansion;

Step 3, using the detail layer image mapping function to stretch the detail layer image to obtain the stretched detail layer image;

Step 4: Synthesize the image of the illumination layer after pixel value range expansion and the image of the detail layer after stretching into a new image; and then convert the synthesized image and the H channel and S channel data obtained in step 1 into an output image.

Further, in the step 1, the V channel data is decomposed into the illumination layer image and the detail layer image, and the specific process is as follows:

I(x,y)=c(x,y)×L(x,y) (1)

Among them, I(x,y) represents the pixel value of the V channel data at the pixel point (x,y), L(x,y) represents the pixel value of the illumination layer image at the pixel point (x,y), c( x, y) represents the pixel value of the detail layer image at the pixel point (x, y).

Further, the illumination layer image is calculated by convolving the V channel data with a Gaussian kernel with a variance of 1.

Further, the concrete process of described step 2 is:

Step 21: Normalize the illumination layer image to obtain a normalized illumination layer image;

代表归一化处理后的光照层图像在像素点(x,y)处的像素值，max表示取最大值运算；in,

Represents the pixel value of the normalized illumination layer image at the pixel point (x, y), and max represents the operation of taking the maximum value;

Step 22: Use the maximum inter-class variance method to determine the brightness segmentation threshold T of the normalized illumination layer image;

Steps 2 and 3: Perform global luminance mapping on the normalized illumination layer image according to the luminance segmentation threshold T and the global luminance mapping function to obtain an illumination layer image after the global luminance mapping;

The global luminance mapping function is:

代表全局亮度映射后的光照层图像在像素点(x,y)处的像素值；in,

Represents the pixel value at the pixel point (x, y) of the illumination layer image after global luminance mapping;

Step 24: Expand the pixel value range of the illumination layer image after global brightness mapping:

Among them, L _d (x, y) represents the pixel value at the pixel point (x, y) of the illumination layer image after the pixel value range is extended.

Further, the detail layer image mapping function is:

Among them, e is the base of the natural logarithm, A, B, D are the coefficients of the detail layer image mapping function, and S(c(x,y)) represents the pixel point (x,y) in the stretched detail layer image. value.

Further, the coefficients A, B and D of the detail layer image mapping function are:

Among them, [h ₀ , h ₁ ] is the definition domain of the pixel value c(x, y) in the detail layer image, c(x,y)∈[h ₀ ,h ₁ ], that is, h ₀ is the pixel value in the detail layer image The minimum value of the value c(x, y), and h ₁ is the maximum value of the pixel value c(x, y) in the detail layer image.

Further, in the fourth step, the image of the illumination layer after the pixel value range has been expanded and the image of the detail layer after being stretched are synthesized into a new image, and the specific process is as follows:

Multiply the pixel values in the same position in the image of the illumination layer after the pixel value range is expanded and the image of the detail layer after being stretched, and use the multiplication result as the pixel value at the corresponding position in the synthesized new image.

The beneficial effects of the present invention are:

In order to improve the overall contrast of the low-illumination image and enhance the details and texture characteristics of the low-illumination image, the present invention proposes a low-illumination image quality improvement method based on a Retinex model. The invention firstly obtains the detail layer image and the illumination layer image by layering the digital image through the Retinex model; secondly, a nonlinear global brightness mapping function is designed to map the illumination layer image to obtain the illumination layer enhanced image; The detail layer image mapping function stretches the detail layer image to obtain the detail layer enhanced image; finally, each pixel of the detail layer enhanced image and the illumination layer enhanced image is multiplied to synthesize the low illumination enhanced image. The experimental results show that the design algorithm of the present invention can effectively improve the image quality of low-illumination images.

Description of drawings

1 is a flowchart of a method for improving low-light image quality based on a Retinex model according to the method of the present invention;

FIG. 2 is a graph of the corresponding global luminance mapping function of the illumination layer when the luminance segmentation threshold T=0.2;

FIG. 3 is a graph of the global luminance mapping function of the corresponding illumination layer when the luminance segmentation threshold T=0.5;

Fig. 4 is the graph of detail layer image mapping function;

Among them, the intensity value range of the original image detail layer is [0.1, 3];

Figure 5(a) is the original image one;

Fig. 5(b) is the enhanced image corresponding to Fig. 5(a);

Figure 6(a) is the second original image;

Fig. 6(b) is the enhanced image corresponding to Fig. 6(a).

Detailed ways

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS First, the present embodiment will be described with reference to FIG. 1 . A method for improving low-light image quality based on the Retinex model described in this embodiment, the method specifically includes the following steps:

Step 1. After converting the acquired image from the RGB channel to the HSV channel, the data of the V channel, the H channel and the S channel are obtained respectively;

Then decompose the V channel data into the light layer image and the detail layer image;

Step 2: Perform global brightness mapping processing on the illumination layer image to obtain an illumination layer image after global brightness mapping;

Then, extend the pixel value range of the illumination layer image after the global brightness mapping, and obtain the illumination layer image after the pixel value range expansion, that is, the illumination enhancement image;

Step 3, using the detail layer image mapping function to stretch the detail layer image to obtain the stretched detail layer image, that is, the detail enhanced image;

Step 4: Synthesize the image of the illumination layer after pixel value range expansion and the image of the detail layer after stretching into a new image; and then convert the synthesized image and the H channel and S channel data obtained in step 1 into an output image.

In this embodiment, firstly, the image is input to the RGB to HSV module for color conversion; then the Retinex layered model shown in formula (1) is separately performed on the V channel data to obtain the illumination layer image and the detail layer image; Perform global brightness mapping as shown in formula (3) to obtain an enhanced image of the illumination layer; map the detail layer image as shown in formula (5) on the detail layer image separately to obtain an enhanced image of the detail layer; Multiply the enhanced image with the illumination layer pixel by pixel to synthesize a new image; through the HSV to RGB module, convert the synthesized new image, the original H channel and S channel data into an output image that can be directly displayed, and the quality of the obtained output image Significantly improved.

Embodiment 2: The difference between this embodiment and Embodiment 1 is that in step 1, the V channel data is decomposed into an illumination layer image and a detail layer image, and the specific process is:

I(x,y)=c(x,y)×L(x,y) (1)

Among them, I(x,y) represents the pixel value of the V channel data at the pixel point (x,y), L(x,y) represents the pixel value of the illumination layer image at the pixel point (x,y), c( x, y) represents the pixel value of the detail layer image at the pixel point (x, y).

According to the Retinex model, digital images can be decomposed into illumination layer images and detail layer images. The illumination layer image determines the overall light-dark contrast of the image, and the detail layer image determines the details and texture of the image. The pixel point (x, y) is the coordinate in the image coordinate system, with the width direction of the image as the x-axis, and the height direction of the image as the y-axis.

Other steps and parameters are the same as in the first embodiment.

Embodiment 3: The difference between this embodiment and Embodiment 1 or 2 is that the illumination layer image is calculated by convolution of V channel data and a Gaussian kernel with a variance of 1.

Other steps and parameters are the same as in the first or second embodiment.

Embodiment 4: The difference between this embodiment and one of Embodiments 1 to 3 is that the specific process of the second step is:

Step 21: Normalize the illumination layer image to obtain a normalized illumination layer image;

代表归一化处理后的光照层图像在像素点(x,y)处的像素值，max表示取最大值运算；in,

Represents the pixel value of the normalized illumination layer image at the pixel point (x, y), and max represents the operation of taking the maximum value;

Step 22: Use the maximum inter-class variance method (OTSU) to determine the brightness segmentation threshold T of the normalized illumination layer image;

Steps 2 and 3: Perform global luminance mapping on the normalized illumination layer image according to the luminance segmentation threshold T and the global luminance mapping function to obtain an illumination layer image after the global luminance mapping;

The global luminance mapping function is:

代表全局亮度映射后的光照层图像在像素点(x,y)处的像素值；in,

Represents the pixel value at the pixel point (x, y) of the illumination layer image after global luminance mapping;

The overall contrast of the image can be enhanced by global brightness mapping. As can be seen from Figures 2 and 3, the global brightness mapping function stretches the brightness less than the threshold T, thereby improving the brightness of this part of the pixels, while for the brightness greater than the threshold T Compression is performed to reduce the brightness of this part of the pixel.

Step 24: Expand the pixel (brightness) value range of the illumination layer image after global brightness mapping:

Among them, L _d (x, y) represents the pixel value at the pixel point (x, y) of the illumination layer image after the pixel value range is extended.

By stretching the illumination layer image L by the method of this embodiment, the overall light-dark contrast of the low-illumination image can be enhanced.

Other steps and parameters are the same as one of the first to third embodiments.

Embodiment 5: This embodiment is described with reference to FIG. 4 . The difference between this embodiment and one of Embodiments 1 to 4 is that the detail layer image mapping function is:

Among them, e is the base of the natural logarithm, A, B, D are the coefficients of the detail layer image mapping function, and S(c(x,y)) represents the pixel point (x,y) in the stretched detail layer image. value.

The nonlinear detail layer image mapping function designed in this embodiment is to enhance the detail and texture characteristics of the detail layer image, and enrich the detail and texture characteristics of the low-illumination image.

Other steps and parameters are the same as one of the first to fourth embodiments.

Embodiment 6: The difference between this embodiment and one of Embodiments 1 to 5 is that the coefficients A, B, and D of the detail layer image mapping function are:

Among them, [h ₀ , h ₁ ] is the definition domain of the pixel value c(x, y) in the detail layer image, c(x,y)∈[h ₀ ,h ₁ ], that is, h ₀ is the pixel value in the detail layer image The minimum value of the value c(x, y), and h ₁ is the maximum value of the pixel value c(x, y) in the detail layer image.

A, B, and D are three undetermined coefficients, which determine the definition domain, value domain, and original image detail intensity (ie, pixel value) of the function to be 1 corresponding to the enhanced image detail intensity. Since the image is the most blurred when the original detail layer image intensity is 1, the detail enhancement image intensity should also be 1, that is, S(1)=1. In addition, the image enhancement of the detail layer needs to keep the definition domain and the value domain consistent, that is, S(h ₀ )=h ₀ , S(h ₁ )=h ₁ . Therefore, these three correspondences are respectively brought into the function (5) to obtain the values of the three undetermined coefficients as shown in the formula (6).

Other steps and parameters are the same as one of the specific embodiments one to five.

Embodiment 7: The difference between this embodiment and one of Embodiments 1 to 6 is that in step 4, the image of the illumination layer after the pixel value range is expanded and the image of the detail layer after stretching are synthesized into a new image, The specific process is:

Multiply the pixel values in the same position in the image of the illumination layer after the pixel value range is expanded and the image of the detail layer after being stretched, and use the multiplication result as the pixel value at the corresponding position in the synthesized new image.

That is, multiply the pixel value of the pixel point (x, y) in the illumination layer image after the pixel value range is extended by the pixel value of the pixel point (x, y) in the stretched detail layer image, and use the multiplication result as The pixel value of the pixel point (x,y) in the synthesized new image.

Other steps and parameters are the same as one of Embodiments 1 to 6.

Analysis of results

The simulation software used in the present invention is Matlab 2018a. The hardware platform is a desktop computer, and its hardware consists of: i9-11900H specification CPU, 16GB DDR4 specification memory, and RTX 3060 specification graphics card. The input and output of the simulation program are standard images in bmp format.

The simulation results of the method of the present invention are shown in Fig. 5(a) and Fig. 5(b) and Fig. 6(a) and Fig. 6(b).

As can be seen from Figure 5(a) and Figure 5(b) and Figure 6(a) and Figure 6(b), due to the relatively low light intensity and uneven distribution, the original image presents a darker effect, causing the human eye The details and outlines of the scene in the original image cannot be distinguished, so the image quality is poor; the low-illumination image quality improvement algorithm designed in the present invention improves the image brightness in the low-illumination area and enhances the local details of the image to be clearer and richer. Therefore, the algorithm of the present invention can effectively improve the overall visualization effect of the low-illumination image, and the image quality of the enhanced image can be effectively improved.

The above calculation examples of the present invention are only to illustrate the calculation model and calculation process of the present invention in detail, but are not intended to limit the embodiments of the present invention. For those of ordinary skill in the art, on the basis of the above description, other different forms of changes or changes can also be made, and it is impossible to list all the embodiments here. Obvious changes or modifications are still within the scope of the present invention.

Claims (5)

1. A low-illumination image quality improvement method based on a Retinex model is characterized by specifically comprising the following steps of:

step one, after the acquired image is converted from an RGB channel to an HSV channel, data of a V channel, an H channel and an S channel are respectively acquired;

in the first step, the V-channel data is decomposed into an illumination layer image and a detail layer image, and the specific process is as follows:

I(x,y)＝c(x,y)×L(x,y) (1)

wherein, I (x, y) represents a pixel value of the V channel data at the pixel point (x, y), L (x, y) represents a pixel value of the illumination layer image at the pixel point (x, y), and c (x, y) represents a pixel value of the detail layer image at the pixel point (x, y);

decomposing the V channel data into an illumination layer image and a detail layer image;

step two, carrying out global brightness mapping processing on the illumination layer image to obtain an illumination layer image after global brightness mapping;

the specific process of the second step is as follows:

step two, normalizing the illumination layer image to obtain a normalized illumination layer image;

wherein,

representing the pixel value of the illumination layer image after normalization processing at the pixel point (x, y), wherein max represents the maximum value operation;

secondly, determining a brightness segmentation threshold T of the illumination layer image after normalization processing by adopting a maximum inter-class variance method;

step two, according to the brightness segmentation threshold value T and a global brightness mapping function, global brightness mapping is carried out on the illumination layer image after normalization processing, and the illumination layer image after global brightness mapping is obtained;

the global luminance mapping function is:

wherein,

representing a global luminance mapThe pixel value of the subsequent illumination layer image at the pixel point (x, y);

step two, carrying out pixel value domain expansion on the illumination layer image after global brightness mapping:

wherein L is _d (x, y) represents the pixel value of the illumination layer image at the pixel point (x, y) after the pixel value domain expansion;

then, carrying out pixel value domain expansion on the illumination layer image after global brightness mapping to obtain an illumination layer image after pixel value domain expansion;

stretching the detail layer image by using a detail layer image mapping function to obtain a stretched detail layer image;

step four, synthesizing the illumination layer image with the expanded pixel value domain and the stretched detail layer image into a new image; and converting the synthesized image and the H channel and S channel data obtained in the first step into an output image.

2. The method as claimed in claim 1, wherein the illumination layer image is obtained by performing convolution calculation on V-channel data and a gaussian kernel with a variance of 1.

3. A method of improving image quality under low illumination based on Retinex model as claimed in claim 2, wherein the detail layer image mapping function is:

wherein e is the base of the natural logarithm, A, B, D are coefficients of the detail layer image mapping function, and S (c (x, y)) represents the value of the pixel point (x, y) in the stretched detail layer image.

4. The method of claim 3, wherein coefficients A, B, and D of the detail layer image mapping function are:

wherein, [ h ] ₀ ,h ₁ ]For the domain of pixel values c (x, y) in the detail layer image, c (x, y) is E [ h ] ₀ ,h ₁ ]I.e. h ₀ Is the minimum value of the pixel values c (x, y), h, in the detail layer image ₁ Is the maximum value of the pixel value c (x, y) in the detail layer image.

5. The method according to claim 4, wherein in the fourth step, the illumination layer image after pixel value domain expansion and the detail layer image after stretching are synthesized into a new image by a specific process:

and multiplying the pixel values at the same position in the illumination layer image after the pixel value domain expansion and the stretched detail layer image, and taking the multiplication result as the pixel value at the corresponding position in the synthesized new image.

Images