Skip to content
BY 4.0 license Open Access Published by De Gruyter Open Access December 20, 2023

Low-illumination image enhancement with logarithmic tone mapping

  • Changqing Du EMAIL logo , Jingjian Li and Bin Yuan
From the journal Open Computer Science

Abstract

For low-illumination video sequences, some existing enhancement algorithms have some problems, such as image over-enhancement, color distortion, and inadequate detail processing. Based on luminance detection, we add logarithmic tone mapping to optimize the existing algorithms. The color space of low-illumination video image is converted from the red, green, blue mode to the hue-saturation-intensity mode, and then, logarithmic tone enhancement is applied to the image. Algorithm in this study has an obvious effect on image luminance enhancement and details processing, which makes the low-illumination video show a clear image with more natural visual effect, thus improving the quality of low-illumination video. This algorithm can avoid the problems of overexposure, color distortion, and blurring of detail processing under low illumination. The infrared spectrum of the object can be captured by infrared detection equipment, and the purpose of image enhancement can be achieved by applying the infrared spectrum of the object.

1 Foreword

In order to obtain high-quality images under complex lighting environment, the hardware equipment of image acquisition is improved and upgraded to adapt to different scenes. With the rapid development of electronic equipment, the quality requirements for the acquired images are also becoming higher and higher. However, there are always many uncontrollable factors in the actual process of image acquisition, which lead to various defects in the obtained images, especially due to the too dark or insufficient exposure around the shooting environment, which will result in loss of image information, image distortion, and darkening etc., resulting in a low-light image. After the image is converted, stored, and transmitted, the quality of such low-illumination images is further deteriorated. Low illumination, as the name suggests, is the normal condition where the ambient illumination does not meet the standard. However, in practical applications, it is impossible to define the low-illumination environment through a specific theoretical value, and there is no unified standard to define the low-illumination category. People regard the images collected by the acquisition equipment in the environment where the ambient light is relatively dark and the intensity of the surrounding light is relatively weak, as low-illumination images. Such low-illumination images have problems such as dark image brightness, low contrast, narrow image grayscale range, easy color distortion, and often contain a lot of noise. These problems seriously affect the subjective visual effect of the image and reduce the application value of the image. Therefore, it needs to be enhanced to improve its visual effect and convert it into a form more suitable for human observation or computer vision system processing. With the rapid development of digital image and multimedia technology and various types of optical imaging system, people for image quality requirements are higher and higher, such as all kinds of monitoring system and target tracking system. But because the environment is uncontrollable, the picture quality is often difficult to meet the requirements too. Especially when in the fog or the rain day, the night or other light conditions are weak, the overall image quality is very poor and the contrast is low, so the visibility is difficult to be satisfied that largely brought a lot of inconvenience to the work. Thus, enhancing low-illumination images is of great significance. In real life, the video quality is often affected by natural conditions and video acquisition methods; especially in the case of low light or uneven illumination at night, the video obtained cannot meet the requirements of practical applications. In order to solve the shortcomings of low-illumination video images, this study conducts in-depth research on video image enhancement technology under low illumination. The form of video is closest to the visual experience of human real life, so people can understand the video content more quickly and clearly, so video is widely used in various fields, such as industrial and agricultural production, medicine, aerospace, and security. The development of video has directly or indirectly changed people’s traditional way of life. Farmers can cultivate crops more conveniently and quickly through the video feedback from agricultural machinery; video feedback from mine exploration equipment can more easily detect the current operation progress; doctors can directly observe the reality of the patient’s inspected part through the video images obtained by medical equipment; the police can catch criminals more quickly through video surveillance; emergency troops can accurately understand the scene of the disaster area through the video sent by the drone. Due to the limitations of the surrounding environment, people often cannot take ideal pictures, and there are many reasons for image degradation, among which the most common are cloudy days or low light. The pictures obtained in this environment have common characteristics: low-contrast, low signal-to-noise ratio, serious loss of details, extremely poor visual effects, etc. Therefore, it is very important to study image enhancement methods.

In a low-illumination environment, the visibility of road environment system becomes worse, and the road condition information obtained by drivers through vision is often inaccurate. Melo believes that HDR video on mobile devices is still in its infancy [1]. Liu et al. proposed a fast and effective-single image dehazing algorithm. First, the algorithm estimates rough transmittance and atmospheric light based on the HSI color space. Then, the rough transmittance is smoothed by the guided filter, and the transmittance of the bright gray area is corrected by the threshold method to generate the final transmittance. The low-illumination image enhancement technology improves the low gray value while suppressing the high gray value and enriches the edge and texture information of the whole or local area in the image. Finally, the restored image is obtained by tone mapping [2]. Sikudová et al. argued that the hue management framework prevents hue and brightness changes [3]. Furuta et al. believed that in volumetric seam carving, i.e., seam carving of 3D cost volumes, the best seam surface can be obtained by the cutting of graphics produced by complex graphic construction. So far, the graph cut algorithm is the only solution for body seam sculpting [4]. Francois and Kerkhof introduced HDR content distribution solutions [5]. El Mezeni and Saranovac thought of ELTM as a flexible tone-mapping operator [6]. Zhou et al. proposed strategy that can achieve the desired large-scale CPILS [7]. Liu et al. believed that supervised feature extraction methods have relatively high performance [8]. Zhao et al. believed that method can improve the visual effect of spatial domain image enhancement [9]. Gandhamal et al. believed that the proposed technique can be used for efficient segmentation of tissue structures in medical images [10]. Image is an important means of conveying information, it can intuitively describe the information in natural scenes, and it is widely used in various fields.

Digital image processing technology is based on the human visual system, the input image processing, to achieve the desired effect. Image enhancement refers to the effective processing of image information to finally achieve the goal of improving its visual effect. For different places or special purposes to highlight the global or local features of the image, the original image is not clear or local features become clearer, or emphasize the information that the relevant people are interested in, enlarge or remove the differences in the image, or remove the redundant information that people do not need, so as to enrich the diverse image information, enhance the visual effect, and the enhanced image will be able to meet the needs of a specific location. It is a very popular subject in scientific research to enhance the color images obtained under low-illumination conditions. One of the purposes of image enhancement research is to improve the color visual effect of the images. For the acquired color image with submerged color features, in order to better highlight the global or partial features of the image, a color image enhancement algorithm needs to be used to greatly enhance the global or partial features of the image, highlight its main features, and suppress noise. Interference, etc., make the image clear and recognizable or make some features more prominent, thereby greatly improving the quality of the color image and improving the overall visual effect of the image, so that the processed image can meet specific needs and applications. Information presented in the form of images and videos is a particularly important part of human life and one of the main ways people interact with information. Today’s world is gradually in a scientific and information age with rapid economic development. With the rapid development of my country’s national economy and society, the research on the history of scientific information technology and its application in modern people’s daily work has played an increasingly important leading role. Under normal circumstances, people obtain the desired images through photographing equipment, and the photographed images are affected by the configuration of the photographing equipment and the shooting environment (light, weather, temperature, etc.). At night or when the lighting conditions are weak or insufficient, the images obtained by people often have the characteristics of low color contrast, poor resolution, small dynamic range, large noise, and poor visual effects. It is difficult to obtain the detailed information of the image, and the visual effect is not ideal for people. Compared with grayscale images and binary images, color images have more color information and are more in line with the characteristics of objects that people see in daily life. Therefore, the research on color image processing is more in line with people’s daily needs, but it also makes the image the complexity of the research has increased. Today’s society is in an era of highly developed information technology. According to scientific statistics, 70 to 80% of people’s information is obtained through the visual system. As the basis of visual information, image has the characteristics of intuition and rich content. It has become an important way for people to acquire, record, and transmit information. Therefore, more and more scholars have entered the research field of image processing. With the continuous development and progress of digital image processing technology, its applications are becoming more and more extensive, such as video surveillance, medical imaging, traffic safety, satellite remote sensing, and other fields, which play a very important role in the development and progress of society. However, in the process of image acquisition, it is difficult to obtain high-quality images due to factors such as acquisition equipment, external conditions, and imaging principles. For example, insufficient light or insufficient exposure caused by nighttime surveillance or other extreme weather results in a significant decrease in the quality of the collected images, which makes the image information that people need lost, and low-light images are also generated. It can be seen that it is inevitable to obtain low-light images, and it is more important to process low-light images. In order to solve this problem, people use digital image processing technology to process the collected images, so as to improve the image quality and meet the needs of people to obtain information. Image enhancement technology is currently a research hotspot in the field of image processing. Its purpose is to improve the visual effect of the image, enhance or suppress the whole or part of the image in a targeted manner, improve image quality, enrich image details, and satisfy people’s needs. Enhancing low-illumination images can improve image brightness and contrast, highlight image detail information, and ensure image color information. To address the global histogram equalization problem, different histogram transformation functions are designed according to different grayscale distributions in each pixel neighborhood of the image. This method is called local histogram equalization. This method selects the image block to be processed in the image to be processed and moves the center of the area from one pixel to another pixel. This method needs to traverse each pixel of the image and perform histogram equalization in the neighborhood of each pixel. Using this processing, the image can display more detailed texture information and achieve better visual effects, but the calculation of this method is more complicated and the operation time is longer. In order to obtain low-illumination video images with good visual effects, many researchers have proposed a series of algorithms, but these algorithms often have good results in only one aspect and cannot take into account all aspects. Therefore, the problem of low-light image enhancement needs to be solved urgently in the field of computer vision application and digital image processing. So this study analyzes the characteristics of low-illumination video image, summarizes the principle of luminance enhancement, and calculates the mapping function model using the fitting algorithm, optimizes the luminance detection algorithm, so as to automatically detect the luminance of the image, adaptively adjust the enhancement level, and avoid excessive enhancement effect and reduce the time of real-time video processing, thus resulting in excessive exposure, which does not make the image bright and dark, which is not natural enough and affects the look and feel. Finally, the maximum efficiency can be achieved. The main research of this study is as follows:

  1. Introduction part. On the basis of luminance detection, logarithmic tone mapping is added to optimize the existing algorithm. Convert the color space of low-light video images from the red, green, blue (RGB) mode to the HIS mode,

  2. Methods section. The image is then logarithmically enhanced, preserving the details of the image.

  3. Analysis part. Compared with the experimental results, the algorithm in this study has an obvious effect on image brightness enhancement and detail processing, which makes the low-illumination video present a clear image, and the visual effect is more natural, thus improving the quality of the low-illumination video.

2 Algorithm based on the luminance enhancement principle

At present, there are some classic video enhancement algorithms in the field of low-light video enhancement, and many scholars have conducted further research and exploration on these algorithms. Weak light enhancement of single image is an important task, which has many practical applications. Most existing methods use single-image methods. However, none of these algorithms can comprehensively solve the above-mentioned shortcomings of low-light video. There will be a lot of noise in the video shot under low illumination, which will affect the expression of video details. Since the low-illumination video image has regenerated noise in addition to its own noise, it is necessary to repeatedly denoise the low-illumination video image. However, in this process, the current video image denoising method cannot ensure that the image noise is removed and the image details are kept under low loss. The details of low-light video images are usually very weak. When the video image is enhanced in other aspects, it is easy to lose the detail information of the image. Information is highlighted, ignoring the details of the transition part of the video image, making the image unnatural. Low-light video images are prone to wide dynamic phenomena, which means that the same frame of video image contains both high-brightness areas and dark areas, so when the brightness of the dark areas is enhanced, it will lead to over-exposure of the high-brightness areas. The video image cannot solve the brightness enhancement problem of the wide dynamic image well, and the brightness gradient of the original video image is ignored in the process of image brightness enhancement, so that the enhanced video image loses the brightness level of the original image.

Video images will not only generate noise due to changes in illumination, but also when there are moving objects in the image, the noise in the video image will also increase rapidly. Especially when the illumination is uneven and the moving objects move too fast, the area of the image sensor receiving light will also change rapidly, which will lead to a lot of noise in the generated video images. In addition to the noise generated in the process of video image acquisition, reproduction noise is also generated in the process of video image enhancement. For example, when the Gaussian kernel is applied in the enhancement process, the image will be affected by Gaussian noise (the so-called Gaussian noise refers to a class of noise whose probability density function obeys Gaussian distribution [i.e., normal distribution]); when the exponential or logarithmic function is applied, some exponential and logarithmic noises will be generated correspondingly. Gaussian noise refers to a class of noise whose probability density function follows a Gaussian distribution (i.e., a normal distribution).

Algorithms based on the brightness enhancement principle are mainly divided into the following categories. The first category is the algorithm based on ton-mapping [11]. This algorithm came into being to overcome the problem of photo distortion when printing photos, it implements some functions to write and transform pixel grayscale, which can effectively enhance the brightness of dark areas of the image. Although this algorithm is relatively simple and its details also are better maintained, the overall contrast is distinctly low. The global method is characterized by its simplicity and speed, and its performance is inferior to that of local algorithm. Pixel points with the same color in the local method may have different colors after mapping, resulting in halo phenomenon (the color of the halo is generally internal infrared purple, color halo), but the performance is better.

Second, it is histogram equalization [12]. This algorithm has fast repair speed, and it of image grayscale range pull liter or let gray uniform distribution, can increase the contrast, make image visibility is higher, whereas the output of the image gray histogram [13] though it is close to uniform distribution; at the same time, its value and the ideal value of 1/n is still likely to exist bigger difference so as to cannot get the best value, and the image exposure is too large to affect the look and feel [14].

The third one is the retinex algorithm, which is an algorithm based on color consistency or color constancy. Its theoretical construction is mainly based on three foundations [15]: its basic content is mainly the object’s color, which is determined by the object’s reflection ability of red long wave, green medium wave, and blue short wave light, so the color of the object is not affected by uniform illumination; with consistency, its advantage can effectively deal with the distortion of color, but it is easy to appear halo phenomenon. Due to the deviation between ideal and reality, reality and survey, it needs to reflect the real performance of the correction algorithm as much as possible.

Image is an important means of conveying information, it can intuitively describe the information in natural scenes and is widely used in various fields. High-quality images not only provide people with better visual effects, but also provide better research conditions for computer fields such as object detection and behavior recognition. Therefore, scholars have continuously improved the hardware equipment and software algorithms to improve the image quality. In order to obtain high-quality images in complex lighting environments, people have improved and upgraded the hardware equipment for image acquisition to adapt to different scenes. However, high equipment manufacturing cost and complicated operation make it unable to be widely used. Compared with the limitation of hardware, digital image processing technology has the characteristics of strong flexibility and easy operation, which effectively makes up for the defects of hardware technology, and image enhancement is one of the important technologies. One is the development of computer, the second is the development of mathematics, and the third is the growth of a wide range of agricultural and animal husbandry, forestry, environment, military, industrial, and medical applications. Image enhancement technology adjusts the gray level of the image purposefully to highlight the overall structural features or local detail features in the scene, so as to enrich the amount of image information and improve the visual effect. An image represented by grayscale is called a grayscale image to form a complete image, which is composed of three channels of red, green, and blue. This technology solves the problem of low dynamic range of hardware device sensors, makes the image have high contrast and good visual effect, and provides richer image information for subsequent tasks such as feature extraction and target recognition.

3 Optimization of low-illumination image enhancement algorithm based on luminance detection

Aiming at the shortcomings of the existing low-illuminance image enhancement methods, we optimize the algorithm based on luminance detection, i.e., on the basis of luminance detection and the low-illuminance image, logarithmic tone mapping is automatically realized. The low-light image is formed under the condition of insufficient illumination. The overall brightness of the image is very low, and the pixel values are concentrated in a very narrow dynamic range. As a result, the obtained image contrast is very low and the details are difficult to reflect. Calculate the mean and variance of the image on the grayscale image. If there is an abnormal brightness, the mean will deviate from the reference point. The specific algorithm is as follows.

3.1 Conversion of color space

Convert the color space of the sample to HSI. Because in HSI (it reflects the way that human visual system perceives color, perceiving color by three basic characteristic quantities: hue, saturation, and brightness) space, I represent luminance, and its luminance enhancement effect is better than that of the initial space. Initial color space is generally RGB (RGB color is often referred to as the optical three primary colors. It is called the optical three primary colors because any color that can be seen by the naked eye in nature can be mixed and superimposed by these three colors, so it is also called additive color mode) and YUV, and the formula for RGB to HSI is as follows:

H = θ , G B 2 π θ , G < B , where θ = cos 1 ( R G ) + ( R B ) 2 ( R G ) 2 + ( R B ) ( G B )

S = 1 3 min ( R , G , B ) R + G + B ,

I = R + G + B 3 .

Another problem of low-illumination images is that there is a lot of interference information in the darker areas. The image obtained under dim conditions has the characteristics of low gray level, not obvious information and high noise content. The enhancement algorithm will amplify the interference information when improving the image contrast, which will reduce the quality of the enhanced image. To solve this problem, scholars have proposed image enhancement methods based on image decomposition. This kind of method first decomposes the image into two parts: illumination component and reflection component according to the Retinex theory (the color of an object is determined by its ability to reflect long wave, medium wave, and short wave light, rather than by the absolute value of reflected light intensity; the color of the object is not affected by the heterogeneity of lighting and has consistency). Retinex is mainly used to solve problems such as light imbalance and color deviation in digital images. It is also widely used in image processing tasks such as haze images and underwater images to obtain high-contrast images. Among them, the illumination component has low-frequency characteristics, which represent the illuminance information and structural information of the image; the reflection component has high-frequency characteristics, which represents the details and noise of the image. After the two components are obtained, they are processed separately and synthesized to obtain the final enhancement result. This kind of method separates the interference information from the brightness information and avoids magnifying it when improving the contrast of the image. Scholars have proposed many image enhancement methods based on image decomposition. YUV color space can be first converted into RGB space and then converted into HSI using the aforementioned formula, where YUV (is a color coding algorithm) to RGB formula is as follows:

R G B = 1 0 1.402 1 0.34414 0.71414 1 1.1772 0 Y U V .

3.2 Establish the mapping relationship between image luminance and control parameters of enhancement function

Each image in the sample data set is enhanced by logarithmic luminance conversion function. The conversion function is as follows:

I en_g ( y ) = log I ( y ) + 1 I max + 1 log 10 log 2 + I ( y ) I max log b log 0.5 8 I max ,

where I ( y ) represents the pixel value of pixel point Y in channel R, channel G, or channel B before global contrast stretching; I max is the maximum pixel value of all pixels in channel R, channel G, or channel B; I en_g ( y ) is the pixel value corresponding to the current pixel point Y after global contrast stretching; and B is the constant controlling the contrast stretch degree.

Using curve spline-fitting method, the function relation between ( b , f ) is established as the mapping function model of adaptive control of luminance enhancement parameters. The following equation is obtained:

a 0 n + a 1 i = 1 n f i + + a k i = 1 n f i k a 0 i = 1 n f i + a 1 i = 1 n f i 2 + + a k i = 1 n f i k + 1 a 0 i = 1 n f i k + a 1 i = 1 n f i k + 1 + + a k i = 1 n f i 2 k

Expressing these equations in the matrix form yields the following matrix:

n i = 1 n f i i = 1 n f i k i = 1 n f i i = 1 n f i 2 i = 1 n f i k + 1 i = 1 n f i k i = 1 n f i k + 1 i = 1 n f i 2 k a 0 a 1 a k = i = 1 n b i i = 1 n f i b i i = 1 n f i k b i .

By solving the aforementioned matrix, we obtain the coefficient matrix and the fitting curve.

4 Experimental results and algorithm analysis

Low-illuminance image is improved based on luminance detection, and the logarithmic tone mapping of low-illuminance image is automatically realized, i.e., the enhancement level suitable for local image is automatically adjusted. The purpose of hue mapping algorithm is to solve the mismatch between high dynamic range image and display device. Improved algorithm greatly improves the image quality, properly handles the image details, avoids the problem of global exposure, and has better visual effect.

4.1 Video enhancement block diagram

The flow chart of program optimization algorithm is shown in Figure 1. First, you need to set the threshold range Y and then perform discriminant operations on the threshold. If it is less than the minimum value, the brightness enhancement is carried out, if it is greater than the high radiation range, in this process, a color enhancer is needed, and so on until the end. System could determine the luminance value of the video on condition that the external environment is dark, and the system enhances the luminance by feedback adjustment; on the contrary, the luminance of the video itself is high, and no feedback is given. According to the luminance component value after the video conversion, the enhancement threshold is set to avoid the use of the enhancement algorithm in the environment of unnecessary enhancement. When the luminance component Y is less than the minimum value, the histogram algorithm is used to enhance the luminance and return to continue judgment. Chrominace enhancement adjustment is carried out after the luminance reaches the set threshold. If the value of external luminance is within the normal range, the chrominace is enhanced to increase the recognition effect. Low-illuminance image enhancement is an important technique to improve image quality by enhancing the overall or local contrast of the image, which can effectively improve the visual effect of the image.

Figure 1 
                  Flow chart of video enhancement algorithm.
Figure 1

Flow chart of video enhancement algorithm.

Severe haze weather not only causes harm to human health, but also has strong attenuation, refraction, and scattering effects on light, forming low-illumination images, which seriously degrades image quality, loses a lot of detail information, and obtains images with poor visual effects. In a low-light environment, the visibility of the road environment system becomes poor, the road condition information obtained by the driver through vision is often inaccurate, and traffic accidents are very likely to occur. What is more dangerous is that driving at night, the driver cannot see the front and the surrounding clearly. If there is a lack of judgment on road information, the traffic accident rate will increase significantly. At the same time, in the low-light environment, the monitoring system cannot correctly identify the target object, and the common monitoring camera is underexposed, resulting in image degradation and many low-contrast dark connected areas, which make the image lose a lot of detail information. In addition to the aforementioned systems, the low-light environment results in the lack of information in parts of the images with poor visual effects, which brings great difficulties in determining the target, directly restricting and affecting the effectiveness of outdoor target recognition and tracking, military reconnaissance, remote sensing imaging, and other systems, resulting in extremely serious property and life damage.

The low-light image is formed under the condition of insufficient illumination, the overall brightness of the image is very low, and the pixel values are concentrated in a very narrow dynamic range, which makes the obtained image contrast very low and the details are difficult to reflect. At the same time, the image captured by the camera under the condition of insufficient brightness will be interfered by various noises, which makes the image signal-to-noise ratio very low, and it is more obvious in the dark area. The existing low-light image enhancement methods can be divided into non-physical model methods and physical model-based methods according to whether the physical model is used, and they can be divided into single-image-based processing methods and multiple-image-based processing methods according to whether only a single image is processed.

With the rapid development of modern society, people’s demand for information is also increasing, and digital images are one of the important sources of information. Therefore, the theories and technologies related to image processing have received extensive attention. High-quality image signals are the guarantee for the efficient work of computer vision systems. However, in real life, often due to insufficient lighting brightness, the light often comes from unnatural light sources, resulting in insufficient light entering the sensor and resulting in the brightness of the collected images. Too low, the contrast is too dark, and a lot of details are invisible. Such images not only fail to meet people’s expectations, but also greatly reduce the application value of images. Affected by these factors, it is difficult for people or computer vision equipment to further analyze and use images. In order to make better use of these severely degraded images, it is inevitable to use techniques such as image enhancement. Enhance the useful information in the image, and it can be a distortion process, the purpose of which is to improve the visual effect of the image, for the application of a given image. Image enhancement technology is the important information for the purposeful reproduction of the image, and it is the preprocessing stage of the image. The main purpose of low-light image enhancement is to make the enhanced image more in line with people’s subjective visual experience, and the image can be easily analyzed and processed by computer vision and other equipment. The quality improvement of low-light images is one of the research hotspots in the field of image quality improvement, especially in the field of computer vision such as urban traffic and surveillance video; low-light image enhancement technology has a bright development prospect. The infrared spectrum of the object can be captured by infrared detection equipment, and the purpose of image enhancement can be achieved by applying the infrared spectrum of the object. At present, infrared radiation technology is used in many fields, and the most important application is the recognition of objects under low-illumination conditions. Since the surface temperature of some objects, especially pedestrians and animals, is generally higher than that of the external environment, this method can greatly improve the recognition ability of these objects in the case of insufficient light. But this method has two flaws: first, it only works when the temperature of the object is higher than the surrounding environment, and it fails when the temperature of the object is not much different from the surrounding temperature, such as obstacles such as stones on the road; second, this equipment is generally expensive and inconvenient to operate; so this method has not been widely used.

4.2 Examples and analysis

Example effect tests and comparisons are shown in Figures 2 and 3. It can be seen from Figure 2 that the algorithm in this study has a significant effect on image enhancement, while the effect of other algorithms on image processing is not very obvious. And the original image is an example of a low-illuminance image. The retinex algorithm is based on human visual perception of luminance, and color model proposed a color invariant perception algorithm, the effect of algorithm in terms of the original image is in accord with the actual look of the original image, and it will be closer to the graphics to person visual sense; however, retinex algorithm at present has clear image luminance details. The main purpose of the hue mapping algorithm is to generate “good-looking” images and realize the maximum image contrast; however, the color distortion of the algorithm not only is larger than the original image but also the details are not prominent, so it cannot present a more natural and clear image. Histogram equalization algorithm is usually used to increase the global contrast of many images, so the algorithm effect can be obviously seen that the histogram algorithm improves the global luminance of the image, resulting in excessive exposure, which does not make the image bright and dark, which is not natural enough and affects the look and feel. Our optimized algorithm takes both luminance and tone into consideration in enhancement. The algorithm can avoid the problems of excessive exposure, color distortion, and obscure detail processing under low illumination, and it can greatly improve the visual effect. Image enhancement can effectively stretch the contrast of the image, improve the visual effect of the image, and recover part of the lost details. Therefore, it is also a research direction to introduce lightweight network structure into image enhancement field. The low-illumination image technology proposed in this study should be more applied to enhance military, industrial, agricultural, and other fields.

Figure 2 
                  Comparison between the different algorithm with evening background. (a) Original picture, (b) retinex algorithm, (c) tone mapping algorithm, (d) histogram equalization algorithm, and (e) our algorithm.
Figure 2

Comparison between the different algorithm with evening background. (a) Original picture, (b) retinex algorithm, (c) tone mapping algorithm, (d) histogram equalization algorithm, and (e) our algorithm.

Figure 3 
                  Comparison between the different algorithm with sunset background. (a) Original picture, (b) retinex algorithm, (c) tone mapping algorithm, (d) histogram equalization algorithm, and (e) our algorithm.
Figure 3

Comparison between the different algorithm with sunset background. (a) Original picture, (b) retinex algorithm, (c) tone mapping algorithm, (d) histogram equalization algorithm, and (e) our algorithm.

With the development of image sensing technology and information fusion technology, multi-sensor fusion technology has been more and more widely used. The research of image enhancement algorithm has important theoretical and practical significance in the field of image processing. The work performed in this article still leaves much room for improvement. Under low light conditions, the use of infrared sensors and low light level sensors can make up for the defects of human biological vision. In this study, an image fusion and enhancement system under low light conditions is designed and implemented. On this system, image enhancement and image fusion technologies under low light conditions are studied. The image enhancement, image fusion, and fusion effect evaluation need to be further studied in the future.

  1. Funding information: This study was funded by 2022 Department of Education Research Fund Project: network lightweight video streaming application based on super-resolution reconstruction (Project No.: 2107141037x614).

  2. Conflict of interest: These are no potential competing interests in our article. And all authors have seen the manuscript and approved to submit to your journal. We confirm that the content of the manuscript has not been published or submitted for publication elsewhere.

  3. Data availability statement: The datasets generated during and/or analyzed during the current study are not publicly available due to sensitivity and data use agreement.

References

[1] M. Melo, L. Barbosa, and M. Bessa, “Context-aware HDR video distribution for mobile devices,” Multimed. Tools Appl., vol. 76, no. 15, pp. 16605–16623, 2017.10.1007/s11042-016-3940-ySearch in Google Scholar

[2] J. P. Liu, B. K. Huang, and G. Wei, “A fast effective single image dehazing algorithm,” Tien Tzu Hsueh Pao/Acta Electronica Sin, vol. 45, no. 8, pp. 1896–1901, 2017.Search in Google Scholar

[3] E. Sikudová, T. Pouli, and A. Artusi, “A gamut mapping framework for color-accurate reproduction of HDR images,” IEEE Comput. Graph. Appl., vol. 36, no. 4, pp. 78–90, 2017.10.1109/MCG.2015.116Search in Google Scholar PubMed

[4] R. Furuta, I. Tsubaki, and T. Yamasaki, “Fast volume seam carving with multipass dynamic programming,” IEEE Trans. Circuits Syst. Video Technol., vol. 28, no. 99, pp. 1087–1101, 2018.10.1109/TCSVT.2016.2620563Search in Google Scholar

[5] E. Francois and L. Kerkhof, “A single-layer HDR video coding framework with SDR compatibility,” SMPTE Motion Imaging J. vol. 126, no. 3, pp. 16–22, 2017.10.5594/JMI.2017.2660618Search in Google Scholar

[6] D. M. El Mezeni and L. V. Saranovac, “Enhanced local tone mapping for detail preserving reproduction of high dynamic range images,” J. Vis. Commun. Image Representation, vol. 53, no. MAY, pp. 122–133, 2018.10.1016/j.jvcir.2018.03.007Search in Google Scholar

[7] L. Zhou, G. L. Bai, and X. Guo, “Light beam shaping for collimated emission from white organic light-emitting diodes using customized lenticular microlens arrays structure,” Appl. Phys. Lett., vol. 112, no. 20, pp. 201902.1–201902.5, 2018.10.1063/1.5026836Search in Google Scholar

[8] H. Liu, D. Zhu, and S. Yang, “Semisupervised feature extraction with neighborhood constraints for polarimetric SAR classification,” IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens., vol. 9, no. 7, pp. 1–15, 2017.10.1109/JSTARS.2016.2532922Search in Google Scholar

[9] L. Zhao, A. Wang, and B. Wang, “Image enhancement algorithm based on sub-image fusion,” Syst. Eng. Electron., vol. 39, no. 12, pp. 2840–2848, 2017.Search in Google Scholar

[10] A. Gandhamal, S. Talbar, and S. Gajre, “Local gray level S-curve transformation – A generalized contrast enhancement technique for medical images,” Comput. Biol. Med., vol. 83, no. Complete, pp. 120–133, 2017.10.1016/j.compbiomed.2017.03.001Search in Google Scholar PubMed

[11] C. Lu, S. Wang, and V. Maids, “Fault severity recognition of aviation piston pump based on feature extraction of EEMD paving and optimized support vector regression model,” Aerosp. Sci. Technol., vol. 67, no. aug, pp. 105–117, 2017.10.1016/j.ast.2017.03.039Search in Google Scholar

[12] S. Varvaressos, K. Lavoie, and S. Gaboury, “Automated bug finding in video games,” Comput. Entertain., vol. 15, no. 1, pp. 1–28, 2017.10.1145/2700529Search in Google Scholar

[13] S. Yagyu and H. Takagi, “Queueing model with input of MPEG frame sequences and interfering traffic,” J. Oper. Res. Soc. Jpn., vol. 3, no. 3, pp. 317–338, 2017.10.15807/jorsj.45.317Search in Google Scholar

[14] M. Yazdi and T. Bouwmans, “New trends on moving object detection in video images captured by a moving camera: A survey,” Comput. Sci. Rev., vol. 28, no. MAY, pp. 157–177, 2018.10.1016/j.cosrev.2018.03.001Search in Google Scholar

[15] A. Ullah, J. Ahmad, and K. Muhammad, “Action recognition in video sequences using deep bi-directional LSTM with CNN features,” IEEE Access, vol. 6, no. 99, pp. 1155–1166, 2018.10.1109/ACCESS.2017.2778011Search in Google Scholar

Received: 2022-11-26
Revised: 2023-02-26
Accepted: 2023-03-23
Published Online: 2023-12-20

© 2023 the author(s), published by De Gruyter

This work is licensed under the Creative Commons Attribution 4.0 International License.

Downloaded on 3.3.2024 from https://www.degruyter.com/document/doi/10.1515/comp-2022-0274/html
Scroll to top button