Low illumination image enhancement method in GIS cavity environment

Abstract

The present invention discloses a low-illumination image enhancement method in a GIS cavity environment, which enhances low-frequency images based on an improved illumination map and enhances high-frequency images based on homomorphic filtering, thereby enhancing the brightness of dark areas of the image while highlighting details, making the illumination more uniform, and obtaining an image with better color fidelity.

Description

Technical Field

[0001] The present invention belongs to the technical field of machine vision, relates to image enhancement processing, and in particular to a low-light image enhancement method in a gas insulated switchgear (GIS) cavity environment. Background Art

[0002] Gas Insulated Switchgear (GIS) equipment is a substation device in the power system and plays an important role in the power transmission process. GIS equipment is mainly composed of metal cavity, support insulators, mutual inductors and connectors. However, with the long-term operation of GIS equipment, ablation defects and some foreign matter are prone to occur inside the GIS equipment, thus affecting the normal operation of the equipment.

[0003] In recent years, some scholars have achieved remarkable results in GIS cavity maintenance by using GIS cavity inspection robots equipped with machine vision detection technology. However, the performance of existing defect detection algorithms is affected by the imaging effect of the equipment, especially in closed and low-light environments such as GIS cavities. Technologies such as histogram equalization increase image brightness and contrast by evenly distributing pixels of different brightness, but this will largely lead to excessive image enhancement. This is because the images collected in the GIS cavity environment have problems such as uneven lighting, a single background with dark areas, and lack of prominent details; and traditional image enhancement methods only consider the characteristics of the image at a certain frequency for processing, and the image recovery is poor. Summary of the invention

[0004] In view of the above-mentioned deficiencies in the prior art, the purpose of the present invention is to provide a low-illumination image enhancement method in a GIS cavity environment, which combines an improved homomorphic high-pass filtering processing technology and an improved illumination map estimation method to achieve effective enhancement of low-illumination images.

[0005] In order to achieve the above-mentioned object of the invention, the present invention adopts the following technical solutions to achieve it.

[0006] The low-illumination image enhancement method in a GIS cavity environment provided by the present invention comprises the following steps:

[0007] S1 enhances the low-frequency image based on the improved illumination map estimation to obtain a first enhanced image, including the following sub-steps:

[0008] S11 denoises the original image to generate a low-frequency denoised image;

[0009] S12 estimates the illumination map of the low-frequency denoised image;

[0010] S13 divides the low-frequency denoised image into three-channel images of R, G, and B, and determines the real image of each channel based on the illumination map obtained in step S12;

[0011] S14 combines the real images of the three channels to obtain a first enhanced image;

[0012] S2 enhances the high-frequency image based on homomorphic filtering to obtain a second enhanced image, including the following steps:

[0013] S21 converts the original image from RGB space to YUV space to obtain the brightness component Y and components U and V;

[0014] S22 processes the brightness component Y based on the homomorphic filtering transfer function to obtain a brightness component Y′ after homomorphic filtering;

[0015] S23 combines the brightness components Y′, U, and V, and then converts the combined image from the YUV space to the RGB space to obtain a second enhanced image;

[0016] S3 performs fusion processing on the first enhanced image and the second enhanced image to obtain a final enhanced image.

[0017] In the above step S1, since the illumination of the collected image in the GIS cavity environment is weak and the feature points are sparse, it is necessary to enhance the local brightness of the image. The present invention uses an algorithm based on an improved illumination map to estimate the low-frequency image. The algorithm first removes part of the noise through low-pass filtering to generate a low-frequency image, and then removes the estimated illumination map to achieve the purpose of image enhancement.

[0018] In the above step S11, the original image is denoised by using low-pass filtering to generate a low-frequency denoised image.

[0019] In the above step S12, the low light enhancement algorithm model is modified based on the Retinex model, as shown in the following formula (1):

[0020]

[0021] In the formula, S(x) represents the low-frequency denoised image, R(x) represents the real image to be obtained, T(x) represents the illumination map, the symbol · represents the element-by-element multiplication of the image, and x represents the pixel. For the above formula, the purpose is to obtain the real image R, so as long as the illumination map T is estimated, the value of R can be obtained.

[0022] First, normalize the low-frequency denoised image S, and then find the maximum value of the three channels R, G, and B for each pixel x in S to ensure a rough estimate of the illumination map There will be no oversaturation, as shown in the following formula:

[0023]

[0024] Secondly, the roughly estimated illumination map is filtered by Gaussian filtering. In order to reduce the setting of parameters and achieve better enhancement effect, the improved estimated illumination map is set as:

[0025]

[0026] In the above formula, is the final estimated illumination map, g is the illumination factor, which controls the image enhancement effect. When g is too small, the image brightness is insufficient, and when g is too large, the image color will be too unnatural. Through filter screening, each pixel in the illumination map is controlled between [0,1].

[0027] In the above step S13, referring to the LIME algorithm, the low-light image in the cavity environment is inverted, and the above equations (1) and (3) are combined to obtain:

[0028]

[0029] In the above formula, t represents the global illumination coefficient, which affects the overall brightness of the image. By analyzing the three channels of the image in formula (4), we can get:

[0030]

[0031] Where i represents R, G, and B channels.

[0032] Therefore, the low-frequency denoised image is divided into three-channel images of R, G, and B, where the three channels share the same illumination map Using formula (5), we can calculate the three-channel real image R i (x), and then filter it to make R i Each pixel in (x) is controlled between [0,1], and finally the R, G, and B three-channel images are combined into an enhanced image of the final pixel, that is, the first enhanced image.

[0033] In the above step S2, since the image enhancement technology studied by the present invention is applied to the cavity, it is necessary to study the problems such as uneven image illumination, insufficient details and too few feature points. The image enhancement algorithm based on improved illumination map estimation studied in step S1 can improve the image brightness, but the details still need to be improved and the illumination uniformity needs to be improved. For this reason, step S2 provides a high-frequency image enhancement algorithm based on homomorphic filtering, which obtains a high-frequency image by processing the image brightness component Y through homomorphic high-pass filtering, so as to increase the image details, eliminate uneven illumination and avoid changes in the color of the enhanced image.

[0034] In the above step S21, since the three components of the color image RGB are closely related, image enhancement will destroy the correlation of the three components of RGB, resulting in a decrease in color fidelity. Therefore, only the brightness space Y that does not cause the image color to change is enhanced. The present invention obtains the white light brightness Y under the PAL system, and the relationship between the RGB space and the YUV space is shown in formula (6):

[0035]

[0036] The final extracted brightness component Y=0.222R+0.707G+0.071B, and the U component and V component can be obtained at the same time, U=-0.109R-0.349G+0.458B, V=0.682R-0.620G-0.062B.

[0037] In the above step S22, it can be known from the principle of homomorphic filtering that the selected value of the filter transfer function H(u,v) determines the entire filtering effect, thereby affecting the image enhancement effect. Commonly used filter transfer functions include Gaussian, Butterworth and exponential types. The Butterworth transfer function has the best effect on increasing image details and contrast, so this section uses the Butterworth transfer function, as shown in the following formula:

[0038]

[0039] In formula (7), R H and R L Represent high-frequency and low-frequency gains respectively, c represents a constant, D 0 represents the cut-off frequency, D(u,v) represents the distance from point (u,v) to the filter center (u 0 ,v 0 ), the expression is as follows:

[0040]

[0041] It can be seen that the control parameters of formula (7) are too many. In order to reduce the parameter settings without affecting the filtering effect, the Butterworth transfer function in the embodiment is designed as:

[0042]

[0043] In the formula, p represents the filtering factor, and the filtering effect can be controlled by controlling the value of p.

[0044] The brightness component Y is processed according to the selected homomorphic filtering transfer function to obtain a brightness component Y′ after homomorphic high-pass filtering.

[0045] In the above step S23, the component Y' obtained is combined with the components U and V obtained in step S21. Then, according to the inverse operation of formula (6), the combined image is converted from the YUV space to the RGB space to obtain a second enhanced image.

[0046] Through step S2, the high-frequency information can be improved while suppressing the low-frequency information of the image, and the image details can be made more obvious while ensuring uniform illumination, so that the image contrast is better enhanced, and the obtained enhanced image has good color fidelity.

[0047] In the above step S3, the first enhanced image obtained based on the improved illumination map estimation and the second enhanced image obtained based on the homomorphic high-pass filtering are fused. Here, a linear fusion method is adopted, as shown in the following formula:

[0048] R(x)=α·R r (x)+β·R g (x) (10)

[0049] Where R(x) represents the final enhanced image, R r (x), R g (x) represents the first enhanced image and the second enhanced image respectively, and α and β represent the corresponding linear coefficients respectively.

[0050] Compared with the prior art, the low-illumination image enhancement method in the GIS cavity environment provided by the present invention has the following beneficial effects:

[0051] The present invention enhances low-frequency images based on an improved illumination map, and enhances high-frequency images based on homomorphic filtering, thereby enhancing the brightness of dark areas of the image while highlighting details, making the illumination more uniform, and obtaining an image with better color fidelity. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 A schematic flow chart of a low-illumination image enhancement method in a GIS cavity environment provided by an embodiment of the present invention.

[0053] Figure 2 Schematic diagram of the process of enhancing low-frequency images based on improving illumination map estimation.

[0054] Figure 3 Schematic diagram of the process of enhancing high-frequency images based on homomorphic filtering.

[0055] Figure 4 The enhanced processing results of GIS cavity environment images using different methods; among them, (a) the original image, (b) the processing result using the MSRCR algorithm, (c) the processing result using the LIME algorithm, and (d) the processing result using the method of the present invention. DETAILED DESCRIPTION

[0056] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0057] Example 1: This example provides a method for enhancing low-light images in a GIS cavity environment. Figure 1 As shown, it includes the following steps:

[0058] S1 enhances the low-frequency image based on the improved illumination map estimation to obtain a first enhanced image.

[0059] In this step, since the illumination of the collected image in the GIS cavity environment is weak and the feature points are sparse, it is necessary to enhance the local brightness of the image. The present invention uses an algorithm based on an improved illumination map to estimate the low-frequency image. The algorithm first removes part of the noise through low-pass filtering to generate a low-frequency image, and then removes the estimated illumination map to achieve the purpose of image enhancement. The low-light enhancement algorithm model is based on the Retinex model, as shown in the following formula (1):

[0060] S(x)=R(x)·T(x) (1)

[0061] In the formula, S(x) represents the low-frequency denoised image, R(x) represents the real image to be obtained, T(x) represents the illumination map, the symbol · represents the element-by-element multiplication of the image, and x represents the pixel. For the above formula, the purpose is to obtain the real image R, so as long as the illumination map T is estimated, the value of R can be obtained.

[0062] like Figure 2 As shown, this step includes the following sub-steps:

[0063] S11 denoises the original image to generate a low-frequency denoised image.

[0064] In this step, the original image is denoised using a low-pass filter to generate a low-frequency denoised image.

[0065] S12 estimates the illumination map of the low-frequency denoised image.

[0066] In this step, the low-frequency denoised image S is first normalized, and then the maximum value of the R, G, and B channels of each pixel x in S is found to ensure a rough estimate of the illumination map. There will be no oversaturation, as shown in the following formula:

[0067]

[0068] Secondly, the roughly estimated illumination map is filtered by Gaussian filtering. In order to reduce the setting of parameters and achieve better enhancement effect, the improved estimated illumination map is set as:

[0069]

[0070] In the above formula, is the final estimated illumination map, g is the illumination factor, which controls the image enhancement effect. When g is too small, the image brightness is insufficient, and when g is too large, the image color will be too unnatural. After multiple experimental comparisons, g = 0.6 is set. Finally, the filter is used to filter the image. Each pixel is controlled between [0,1].

[0071] S13 divides the low-frequency denoised image into three-channel images of R, G, and B, and determines the real image of each channel based on the illumination map obtained in step S12.

[0072] Referring to the LIME algorithm, the low-light image in the cavity environment is inverted and combined with equations (1) and (3) above to obtain:

[0073]

[0074] In the above formula, t represents the global illumination coefficient, which affects the overall brightness of the image. After experiments, t=0.99 is taken here. By analyzing the three channels of the image in formula (4), we can get:

[0075]

[0076] Where i represents R, G, and B channels.

[0077] Therefore, the low-frequency denoised image is divided into three-channel images of R, G, and B. Here, the three channels share the same illumination map T%′(x). Formula (5) is used to calculate the three-channel real image R i (x), and then filter it to make R i Each pixel in (x) is controlled between [0,1], and finally the R, G, and B three-channel images are combined into an enhanced image of the final pixel, that is, the first enhanced image.

[0078] S14 combines the real images of the three channels to obtain a first enhanced image.

[0079] The low-frequency image enhancement method based on improved illumination map estimation is used to enhance different low-light images. This method can effectively eliminate the influence of uneven illumination, improve the brightness of dark areas of the image, remove noise and increase contrast. The only drawback is that the details of the enhanced image are not prominent and the brightness is too full. This part of the problem will be dealt with in the next step.

[0080] S2 enhances the high-frequency image based on homomorphic filtering to obtain a second enhanced image.

[0081] Since the image enhancement technology studied in the present invention is applied in the cavity, it is necessary to study the problems of uneven image illumination, insufficient details and too few feature points. The image enhancement algorithm based on improved illumination map estimation studied in step S1 can improve the image brightness, but the details still need to be improved and the illumination uniformity needs to be improved. For this reason, step S2 provides a high-frequency image enhancement algorithm based on homomorphic filtering. The algorithm obtains a high-frequency image by processing the image brightness component Y through homomorphic high-pass filtering, so as to increase the image details, eliminate uneven illumination and avoid changes in the color of the enhanced image.

[0082] like Figure 3 As shown, this step includes the following sub-steps:

[0083] S21 converts the original image from the RGB space to the YUV space to obtain the brightness component Y and the components U and V.

[0084] Since the three components of color image RGB are closely related, image enhancement will destroy the correlation of the three components of RGB, resulting in a decrease in color fidelity. Therefore, only the brightness space Y that does not cause changes in image color is enhanced. The present invention obtains white light brightness Y under the PAL system, and the relationship between RGB space and YUV space is shown in formula (6):

[0085]

[0086] The final extracted brightness component Y=0.222R+0.707G+0.071B, and the U component and V component can be obtained at the same time, U=-0.109R-0.349G+0.458B, V=0.682R-0.620G-0.062B.

[0087] S22 processes the brightness component Y based on the homomorphic filtering transfer function to obtain the brightness component Y′ after homomorphic filtering.

[0088] From the homomorphic filtering principle, it can be known that the selected value of the filter transfer function H(u,v) determines the entire filtering effect, thereby affecting the image enhancement effect. In order to reduce parameter settings without affecting the filtering effect, the Butterworth transfer function in the embodiment is designed as:

[0089]

[0090] Where p represents the filtering factor. The filtering effect can be controlled by controlling the value of p. After many experimental comparisons, p is set to 0.01.

[0091] The brightness component Y is processed according to the selected homomorphic filtering transfer function to obtain a brightness component Y′ after homomorphic high-pass filtering.

[0092] S23 combines the brightness components Y′, U, and V, and then converts the combined image from the YUV space to the RGB space according to the inverse operation of formula (6) to obtain a second enhanced image.

[0093] Through step S2, the high-frequency information can be improved while suppressing the low-frequency information of the image, and the image details can be made more obvious while ensuring uniform illumination, so that the image contrast is better enhanced, and the obtained enhanced image has good color fidelity.

[0094] S3 performs fusion processing on the first enhanced image and the second enhanced image to obtain a final enhanced image.

[0095] In this step, the first enhanced image obtained based on the improved illumination map estimation and the second enhanced image based on the homomorphic high-pass filtering are fused. Here, a linear fusion method is adopted, as shown in the following formula:

[0096] R(x)=α·R r (x)+β·R g (x) (10)

[0097] Where R(x) represents the final enhanced image, R r (x), R g (x) represents the first enhanced image and the second enhanced image, respectively, α and β represent the corresponding linear coefficients, respectively. In the present invention, α and β are 0.7 and 0.3, respectively.

[0098] In order to effectively verify the present invention, images are collected in the actual substation GIS cavity for testing, and the structural similarity (SSIM), peak signal-to-noise ratio (PSNR) and brightness order error (LOE) which are commonly used to evaluate image quality indicators are used as evaluation criteria for image enhancement.

[0099] Table 1 Objective quality evaluation of GIS cavity image enhancement effect

[0100]

[0101] Note: For the MSRCR algorithm, see (Rahman ZU, Jobson DJ, Woodell G A. Retinex processing for automatic image enhancement [J]. Proceedings of SPIE-The International Society for Optical Engineering, 2004, 13: 100-110.);

[0102] For the LIME algorithm, please refer to (Guo

[0103] In order to prove the effectiveness of the present invention, the experimental results of the algorithm are compared with those of the MSRCR algorithm and the classic LIME algorithm, as shown in Table 1 and Figure 4 As shown. It can be seen from the figure that (b) has too much image color restoration and large distortion due to the introduction of color factors, and there is also blurred details; (c) has a significant increase in global brightness, but the visual sense is poor and the color is slightly distorted; (d) enhanced by the method of the present invention effectively compensates for the image brightness, increases the image details, and the image color is normal without distortion. It can be seen from the objective evaluation of the images in Table 1 that compared with the MSRCR algorithm and the LIME algorithm, the PSNR value and SSIM value of the method of the present invention are larger, and the LOE value is smaller, indicating that the enhanced image is less different from the original image, the degree of distortion is smaller, the natural protection of the image is better, and the enhanced image has a better visual effect.

[0104] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0105] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0106] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0107] The present invention uses specific embodiments to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

[0108] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. A low-light image enhancement method in a GIS cavity environment, characterized in that: The following steps are involved: S1 enhances the low-frequency image based on the improved illumination map estimation to obtain a first enhanced image, including the following sub-steps: S11 denoises the original image to generate a low-frequency denoised image; S12 estimates the illumination map of the low-frequency denoised image. First, the low-frequency denoised image S is normalized, and then the maximum value of the three channels of R, G, and B for each pixel x in S is found, as shown in the following formula: Secondly, the roughly estimated illumination map is filtered by Gaussian filtering, and the improved estimated illumination map is set as: In the formula, is the final estimated illumination map, g is the illumination factor; S13 divides the low-frequency denoised image into three-channel images of R, G, and B, and determines the real image of each channel based on the illumination map obtained in step S12; in a specific implementation method, the low-frequency denoised image is divided into three-channel images of R, G, and B, and the three-channel real images R are calculated according to the following formulas i (x): In the formula, i represents R, G, and B channels, and t represents the global illumination coefficient; S14 combines the real images of the three channels to obtain a first enhanced image; S2 enhances the high-frequency image based on homomorphic filtering to obtain a second enhanced image, including the following steps: S21 converts the original image from RGB space to YUV space to obtain the brightness component Y and components U and V; S22 processes the brightness component Y based on the homomorphic filtering transfer function to obtain the brightness component Y′ after homomorphic filtering. The homomorphic filtering transfer function is shown in the following formula: Where p represents the filtering factor; D(u,v) represents the distance from the image midpoint (u,v) to the filter center (u0,v0), and its expression is as follows: S23 combines the brightness components Y′, U, and V, and then converts the combined image from the YUV space to the RGB space to obtain a second enhanced image; S3 performs fusion processing on the first enhanced image and the second enhanced image to obtain a final enhanced image.

2. The low-illumination image enhancement method in a GIS cavity environment according to claim 1, characterized in that: In step S11, the original image is denoised using a low-pass filter to generate a low-frequency denoised image.

3. The low-illumination image enhancement method in a GIS cavity environment according to claim 1, characterized in that: In step S21, the relationship between RGB space and YUV space is as follows:

4. The low-illumination image enhancement method in a GIS cavity environment according to claim 1, characterized in that: In step S3, a linear fusion method is adopted for the first enhanced image and the second enhanced image, as described in the following formula: R(x)=α·R r (x)+β·R g (x) In the formula, R(x) represents the final enhanced image, R r (x),R g (x) represents the first enhanced image and the second enhanced image respectively, and α and β represent the corresponding linear coefficients respectively.

Images