An Adaptive Detail Equalization for Infrared Image Enhancement Based on Multi-Scale Convolution

In order to solve the problem of low contrast and fuzzy detail in infrared image, we propose an infrared image enhancement method based on multi-scale and adaptive bi-interval histogram equalization with details. The method mainly consists of four parts: details enhancement, contrast stretch, edge enhancement and reconstruction of enhancement images. Firstly, the multi-scale convolution is used to enhance the details of image; Secondly, taking maximize the variance between classes and minimize the variance as fitness function and solved the threshold of the infrared image by genetic algorithm, then dividing the infrared image into two sub-intervals according to the threshold. After that, the bi-interval histogram equalization with details is applied to enhance the global contrast, at the same time, using the mean square deviation and average gray equalization to improve the brightness of the image. Finally, the enhanced image by adaptive bi-interval histogram equalization with details and the image processed by adaptive limited Laplace operator are fused by linear weighting to reconstruct the final enhancement image. The experimental results show that the proposed method can outperform state-of-the-art ones in both qualitative and quantitative comparisons.


I. INTRODUCTION
Infrared imaging technology is widely used in aerospace, maritime rescue, and military target detection [1]- [3]. However, the contrast of infrared image is low and the texture details are fuzzy. The main reason is that the infrared ray will be affected by the atmospheric thermal radiation due to the distance between the target and the sensors in the scene is far away [4], [5]. In order to improve the contrast of infrared image and highlight the target details, image enhancement is an effective method [2], [4], [35]. In Fig.1, we give examples of the original infrared images and their enhanced effects in different scenes. The low-quality images with blurred details and low contrast are shown in the top row of Fig.1, which reduce the visual effect of the image and affect people's recognition and perception of the target. The infrared images enhanced by our method are shown in the bottom row The associate editor coordinating the review of this manuscript and approving it for publication was Songwen Pei . of Fig. 1. It can be seen that the enhanced infrared images have higher contrast, clearer texture detail and more comfortable visual effect. Thus, image enhancement has attracted more and more attention from researchers [1]- [5], [36].
In recent years, histogram equalization [6], transform domain equalization [7], [8] and enhancement methods based on image stratification [9] have been widely used for image enhancement and have achieved good enhancement effects. Histogram equalization is an image enhancement method with contrast stretch, which can effectively improve the brightness and contrast of low-quality infrared images. Gautam et al. [10], Dhal et al. [6] and Sim et al. [11] used different thresholds to divide the histogram of the original image into two or more sub-intervals for image equalization operation, so as to improve the image quality. Nevertheless, these methods did not consider the distribution of the original histogram curve when the threshold is selected, especially the enhancement effect was poor when the local peak occurred in the histogram curve of the image. In accordance with image VOLUME 8, 2020 This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ information entropy, Liu et al. [20] and Parihar et al. [12] divided the original image histogram dynamically. Although the method has no parameters and does not need to specify the number of interval division, it needs to calculate the information entropy of the image many times, which increases the complexity of the method. Veluchamy et al. [13] adopted histogram shearing technique, Singh et al. [29], Xiao et al. [15] and Wong et al. [14] adopted frequency weighting technique, Subramani et al. [16] adopted gamma function to modify the enhanced histogram accumulation function can realize image enhancement. These methods are suitable for scenes with a single image target and little fluctuation in the histogram curve, which can subjectively control the degree of image enhancement. Yao et al. [17] and Zhao et al. [18] combined optimization theory with functional analysis theory to enhance the image. These methods did not depend on the original histogram distribution, only needed to solve the target value according to the set conditions. But its mathematical operation was more complex and needed to know the target histogram distribution clearly. Wang et al. [19] and Liu et al. [20] improved the local detail features of the image by partitioning the original image in space and then equalizing each sub-block independently. However, because these methods performed equalization operations on subblocks, which will cause blocky tomographic effects and affect the enhancement effect of the image. In addition, multiple equalizations were required, which increased the complexity of the algorithm. The above methods are mainly based on the image histogram attributes and have achieved good enhancement effects. Therefore, histogram equalization has been widely applied in image defogging [5], medical image processing [26] and target detection [12], [15]. Transform domain equalization is to transform the gray value of image to gradient domain or wavelet domain by gradient transform transformation or wavelet transformation for image enhancement [21], [22]. The gradient domain mainly reflects the details information of the image, and it can display the image gradient value in the gray level as possible. Zhao et al. [23] introduced the gradient domain into the histogram specification and reconstructed the gradient field of the target image by means of weighted fusion of bimodal Gaussian function and original image. In the wavelet domain, the edge and detail features of the image correspond to the high frequency coefficients, and the background and contour information of the image correspond to the low frequency coefficients [21], [24]. Therefore, image enhancement in the wavelet domain is mainly realized by equalization of wavelet coefficients. Aboshosha et al. [25] conducted segmental equalization of medical images in the wavelet domain to obtain high-quality medical images. Sdiri et al. [26] improved the image quality by using the stereo endoscope image enhancement in the wavelet domain. Oulhaj et al. [27] combined the dual-tree complex threshold wavelet transform with other traditional methods to achieve the de-noising and enhancement of medical images, respectively. These methods are mainly used to enhance the details and texture of the image, but not to improve the brightness and overall quality of the image. The enhancement methods based on the idea of image stratification [9] are to divide the original image into detail layer, structure layer or background area and target area, and then process the two layers or areas separately according to the characteristics of each layer or area. Tufai et al. [5] divided the original image into background area and target area for enhancement. Sim et al. [11] and Dhal et al. [6] divided the original image into detail layer and structure layer. These methods could solve the blur problem of background and target area, but could not improve the contrast of image.
All in all, image enhancement can improve infrared image quality and broaden the application of infrared imaging technology. In order to solve the problem of low contrast and fuzzy details of infrared image, we combine multi-scale convolution, genetic algorithm and adaptive bi-interval histogram equalization with details, as well as introduce adaptive limited Laplace operator and linear weighted fusion to propose a novel infrared image enhancement method. Our contributions include three aspects: 1. According to the principle of multi-channel convolution in deep learning, we employ the multi-scale convolution composed of six Gaussian kernel functions with different parameters to enhance image details.
2. In order to improve the contrast and brightness of the image, taking the maximum intra-class variance and the minimum inter-class variance as the fitness function and solving it by genetic algorithm. Then bi-interval histogram equalization with details is used to improve the contrast of the image; At the same time, the mean value and mean square deviation are introduced to improve the brightness of the image.
3. Based on the distribution of image histogram, adaptive limited Laplace operator is proposed to enhance the contour of the target in the image. The rest of this paper is organized as follows: In section II, our method and its detail operation are shown. In section III, the performance of our method in different scenes is evaluated. The conclusion and the exploration of the future work are shown in section IV.

II. METHOD
The description of the proposed image enhancement method in this paper is shown in Fig.2. We divide this method into four stages: multi-scale convolution, adaptive bi-interval histogram equalization with details, adaptive limited Laplacian sharpening and linear weighted fusion. Firstly, in order to preserve detailed information, multi-scale convolution is used to process the original infrared image. Secondly, to improve the brightness and contrast of the image, first of all, we take the maximum intra-class variance and the minimum inter-class variance as the fitness function, and the genetic algorithm is used to solve it to obtain the threshold value; Then, the processed infrared image is divided into two sub-graphs by the threshold value, namely the light layer and the dark layer. Meanwhile, the adaptive bi-interval histogram equalization with details is used to improve the contrast of the image; Finally, the gray level homogenization with mean value and mean square deviation is used to improve image brightness. At the same time, overflow judgment is introduced to ensure that the gray value of the image is between 0 and 255. Thirdly, adaptive limited Laplace operator is used to extract the target contour and detailed features of image C. Fourthly, the image H enhanced by multi-scale convolution and the adaptive bi-interval histogram equalization with details and image D processed by adaptive limited Laplace operator are fused with linear weighting to reconstruct the final enhancement infrared image with clear texture, significant contrast and good visual effect.

A. MULTI-SCALE CONVOLUTION
Deep details of image is extracted by multi-channel convolution in deep learning [28]. Referring to the multi-channel convolution idea of deep learning, six Gaussian convolution kernels with different parameter and sizes of 3 * 3 are used to process low-quality infrared images to extract the detail features of the image from the six scales, and then carry out weighted fusion between the extracted feature sub-graph and the original image, namely: where Image C is the fused image, Image In is the original infrared image, F is the image detail features extracted by multi-scale convolution, β is the weighting factor and set to 0.6. Suppose that the original image is f (x, y), F is obtained by multi-scale convolution as shown in formula: where n is the filtering scale number of the convolution kernel function, * is the convolution operation, and g n (x) is the kernel function of the convolution kernel, which can be specifically expressed as: Gaussian convolution kernel function can adjust the filtering radius of the image by adjusting σ . Multi-scale Gaussian convolution is constructed by setting different filtering radii for image detail feature extraction. The process of extracting infrared image details is shown in Fig.3.

B. ADAPTIVE BI-INTERVAL HISTOGRAM EQUALIZATION WITH DETAILS
Bi-interval histogram equalization [10] uses the mean value of image gray as the threshold to divide the whole gray interval of the image into two sub-intervals. Obviously, the distribution of image histogram is not considered, which has VOLUME 8, 2020 poor enhancement effect for the image with dark brightness or dense histogram distribution. In addition, it has poor effect on enhancing image details. In this paper, the threshold solution of interval division is solved by genetic algorithm (GA). Meanwhile, the detailed information is introduced into the bi-interval histogram equalization. So we propose an adaptive bi-interval histogram equalization with details (ADBHE), which can be divided into two stages: threshold solution and contrast stretch.
Step 1: Threshold solution GA is an intelligent optimization algorithm that simulates biological genetics with population as the research object and can process data in parallel to achieve the purpose of quickly solving the optimal solution of the objective function. The setting of fitness function is very important in the threshold solution of GA. Based on the image gray histogram curve, this paper takes the maximization of intra-class variance and minimization of inter-class variance as the fitness function and uses the image histogram as the variable to evaluate the fitness function. The fitness function is defined as: In (4), V in is intra-class variance and V out is inter-class variance. Where V in is defined as, V out is defined as, where T is the threshold, P T is the probability of segmentation threshold and hist(x) is the histogram within the range of threshold T . The GA realizes the solution of image threshold through four stages: population initialization, fitness function evaluation, replication and termination.
Step 2: Contrast stretch The threshold T solved by GA is used to divide the image into two sub-intervals, namely X = X B ∪ X W , where X B belongs to [min(X ), T ] and X W belongs to [T , max(X )]. Then, the probability density function of the two sub-intervals are defined as, where P B and P W are the probability density functions of sub-intervals X B and X W , n B and n W are the total gray levels of sub-intervals X B and X W , n K B and n K W are the number of gray values X K of sub-intervals X B and X W , respectively.
So the cumulative density function of the two sub-intervals are defined as, where C B and C W are cumulative probability density functions of sub-interval X B and X W , respectively. Due to the loss of detail information in the two subintervals histogram equalization, we introduce the detail information when we use the cumulative probability density function to solve the transformation function, namely (12) where X D B and X D W are detail information of two sub-intervals X B and X W extracted by log operator, f B and f W are the gray value after the transformation of sub-interval X B and X W , respectively. So the final enhanced image gray value Y is In order to improve the overall brightness of the image, the gray value of ImageH is homogenized at [0, 255], i.e where . mean and var are sub-functions to solve the Mean and variance of image gray value, respectively. We can control the degree of image brightness enhancement by adjusting parameter λ. In order to ensure the distribution of image gray value belongs to [0, 255], a judgment condition should be added to Equation (14), namely

C. ADAPTIVE LIMITED LAPLACE OPERATOR
Although the brightness of the enhanced image by ADBHE is significantly improved, the edge information is fuzzy, so Laplacian operator is considered to extract the edge information. Supposing that the gray value of image is M , the gray value of image after sharpening is M , the Laplace operator template is L, and the sharpening process of the image by Laplace operator is defined as: Laplace operator is sensitive to high-contrast regions of image, but it cannot extract the details of low-contrast regions, so an adaptive limited Laplace sharpening operator is proposed based on gradient changes of the image. The basic idea of the algorithm is to reflect the change of image contrast according to the gradient, taking the gradient average as the threshold of the limited Laplace operator, and then using different templates to sharpen the image. Assuming that the gradient of the image is: where dx(i, j) and dy(i, j) are the gradient value in horizontal and vertical direction of the image, respectively. The value is solved by the median difference method, namely: So the threshold T is defined as: Then, we divide the image gray value into two sub-intervals through T , and different templates are used to sharpen the two sub-intervals of the original infrared image. The sharpened Image D is defined as: where f (x) is the gray value of the original infrared image, f (x) is the gray value of the processed infrared image, T is the threshold value of the adaptive limited Laplace operator, L a and L b are Laplace sharpening operator templates. In this paper, L a and L b are respectively defined as:

D. LINEAR FUSION
Image fusion mainly fuses two or more images with low image quality but prominent detail and texture, so as to obtain high-quality images with better visual effect and prominent detail. Referring to the idea of image fusion, this paper fuses Image H with significantly improved brightness and detail retention by ADBHE enhancing and the Image D with edges and details highlighted extracted by the adaptive limited Laplace operator. The fusion result can be describe as: where λ is the weighted coefficient.

III. RESULTS AND ANALYSIS
The proposed method runs on Matlab R2014a in Windows 10 operating system with CPU Intel(R) Core (TM)-i7 9700K 8 Core 3.6GHz and 16G memory. In this paper, infrared images with lower brightness in different scenes are selected for experiments. Firstly, we analyze the effects of multi-scale convolution and adaptive bi-interval histogram equalization with details in this algorithm. Then, the effectiveness of the proposed method is verified by comparative analysis with traditional methods in both quantitative and qualitative aspects. The following is a functional analysis of each operation of the algorithm.

A. MULTI-SCALE CONVOLUTION
Multi-scale convolution is mainly to enhance the details of the infrared image. In this paper, infrared images of buildings and natural environment animals are used to verify the performance of the operation, in which filtering radius of the Gaussian kernel function of multi-scale convolution is set as 0.1, 0.3, 0.5, 0.7, 0.9 and 1 respectively. Fig.4 shows the original image, the enhancement effect of HEEF [32] and multi-scale convolution for the original images of the two scenes. Fig.5 shows the histogram corresponding to the Fig.4 from left to right. From Fig.4, we can see that the images of two scenes are clearer and the quality is improved by HEEF and multi-scale convolution processing. Compared with the original image, the two methods have better enhancement effect, but the quality of two images after multi-scale convolution processing is best. The same conclusion can be drawn from the image histograms shown in Fig.5. Taking infrared images of buildings as an example, it is illustrated that the corresponding histogram of the original infrared images of buildings has almost no distribution after the gray value of about 200, and the gray value of infrared images of buildings also has no distribution after 200 by HEEF processing. However, the gray value of infrared images of buildings still has distribution after 200 by multi-scale convolution processing, indicating the multi-scale convolution can expand the image gray distribution.
In terms of detail enhancement, the red box area in infrared images of buildings and the red box and the orange box area in image of natural environment animals are taken as examples. For the red box area of infrared images of buildings, the dividing line of the building in the original infrared image is fuzzy. By multi-scale convolution processing, grass and animal feet are clearly visible and the image details are more prominent. The experimental results show that multi-scale convolution can enhance the detail and brightness of the image while improving the overall visual effect of the image.

B. ADAPTIVE BI-INTERVAL HISTOGRAM EQUALIZATION WITH DETAILS
When the image is enhanced by bi-interval histogram equalization, the average gray value of the image is used as the threshold value to divide the interval, and the histogram curve distribution of the image is not taken into account, which results in the brightness of the enhanced image is not significantly improved, the details are missing and the contrast is low. However, the adaptive bi-interval histogram equalization with details fully considers the histogram distribution of the image and divides the interval by GA. Two methods, detail retention and brightness enhancement, are used to process the image in two sub-intervals respectively, so that the brightness of the image is improved while the detail information of  image is well maintained. To verify the effectiveness of this method, the image of the building scene was used for the verification experiment and compared with the enhancement effects of HEEF [32] and BBHE [33]. The experimental results are shown in Fig.6 and Table 1.
From the top row in Fig.6, we can see the original infrared image has dark brightness, poor visual effect, and the detail features are not prominent enough. By HEEF [32] processing, the brightness of infrared image is almost unchanged; By BBHE [33] processing, the brightness of infrared image is improved but the local enhancement is not obvious. After adaptive bi-interval histogram equalization with details, the brightness of infrared image is significantly improved. As can be seen from Table 1, the average gray level of the original infrared image is 57, the number of gray levels is 232, and the information entropy is 6.6357. By HEEF [32], BBHE [33] and adaptive bi-interval histogram equalization with details processing, the average gray of the original infrared image are 46, 68 and 88; The number of gray levels is 254, 256 and 256, and the information entropy is 6.5543, 7.2724 and 7.3688, respectively. Obviously, adaptive bi-interval histogram equalization with details can improve the average gray level of the image, that is, improve the brightness of the image.  In terms of detail enhancement, taking the blue box of infrared image as an example, the building contour of the original infrared image and the enhanced image by HEEF [32] and BBHE [33] is blurred. It can be seen from the bottom line of Fig.6 that the lines on the surface of the building are also fuzzy. However, by adaptive doubleinterval histogram equalization with details, the image of the building has clear contour and high contrast. It can be seen from Table.1 that the information entropy of image information decreases by HEEF [32] processing, indicating that HEEF [32] processing will cause the loss of image detail information and reduce the image quality. Through BBHE [33] and adaptive bi-interval histogram equalization with details, the information entropy of image will increase and the latter is higher, indicating that both BBHE [33] and adaptive bi-interval histogram equalization with details can improve the quality of the image and the latter has higher image quality.

C. LINEAR FUSION
Linear fusion mainly makes linear superposition of two or more images with different characteristics to obtain high quality images. However, the weight setting in linear fusion has a great influence on the final fused image. [19], [30] and [34] show that linear fusion has the best effect when the weight is set at 0.5∼0.9. In order to determine the weight λ, the infrared image of the building scene was used to carry out experiments under different λ. The experimental results are shown in Fig.7. It can be seen from the figure that the setting of λ can affect the brightness and detail enhancement of the image. When λ = 0.5 and λ = 0.6, the image brightness is dim. When λ = 0.8 and λ = 0.9, details are lost and obvious overexposure occurs in the image. When λ = 0.7, the brightness of the image is moderate and the details are abundant, so the λ is set as 0.7 in this paper.

D. QUALITATIVE ANALYSIS
In this paper, the enhancement experiment was carried out on three scenes: buildings with obvious edges, people and animals with abundant details and sea surface images with less details. The performance of our method and other five methods were analyzed in both quantitatively and qualitatively. First, we qualitatively analyze the enhancement effect of our method and other five methods. Fig.8 shows the processing results of our method and other five methods for building scenes with obvious edge contours. It can be seen from Fig.8 that DOTHE [29], FCCE [30] and Wan [31] can enhance image brightness, but images processed by DOTHE [29] and FCCE [30] have overexposure and edge blurring, and images processed by Wan [31] have blurring. The enhancement effect of BBHE [33] and HEEF [32] on the image is not obvious, and the brightness is still dim, indicating that the images quality processed by these two methods is still low. In terms of detail enhancement, the yellow boxes in the image A and image B are taken as an example. It can be seen from the second and fourth lines of Fig.8 that the image details processed by DOTHE [29], BBHE [33], FCCE [30] and our method are clear, but the  detail images processed by BBHE [33] are dim. HEEF [32] processed details are still dim and Wan [31] processed details are blurry. However, the windows in the yellow box of image A and the road in the yellow box of image B are moderately exposed and clearly visible with obvious outline by our method processing. Therefore, our method improves the global contrast, improves the brightness and quality of the infrared image. Fig.9 shows the enhanced overall effect and local detail effect of two images of people and animals with rich detail texture features by our method and other five methods. It can be seen from the Fig.9 that the brightness of images processed by DOTHE [29], FCCE [30] and Wan [31] has been improved, but the yellow boxes of image C and image D are still fuzzy after DOTHE [29] and Wan [31] processing, and the contrast of yellow boxes in image C and image D are still low after FCCE [30] processing, especially the goose's feet are still fuzzy in the yellow box of image D. The images processed by HEEF [32] and BBHE [33] are darker with low contrast. For image F, the feather texture is relatively prominent after HEEF [32] processing, but the feet are still not clear, while the goose's feet are still fuzzy and the feather texture is almost non-existent after BBHE [33] processing, and the image details are seriously lost after Wan [31] processing. By our method processing, the overall brightness and contrast of image C and image D are improved, and the visual effect of the image is also improved. In terms of detail enhancement, the human head in the yellow box of image C is processed by our method is obvious, and the goose's feet are highlighted and the feather texture in the yellow box of image D is clear. So our method can also enhance the detail of the image. Fig.10 shows the enhancement effect of a sea boat scene with less details and simple background by our method and other five methods. It can be seen from the Fig.10 that image E and image F by BBHE [33] and DOTHE [29] processing have obvious overexposure; By HEEF [32] processing, the whole image is dark and the contrast is low. Wan [31] can improve the brightness of the image but the contrast of the whole image is low. The brightness of image E and image F is enhanced by our method, at the same time, the contrast of image E and image F are also improved. In terms of detail enhancement, the rope in the yellow box of image E and image F are clearly visible by our method processing. However, the yellow box of image E and image F processed by BBHE [33] and DOTHE [29] have obvious overexposure. Processing by HEEF [32], the yellow box of image E and image F are still dim and the ropes are fuzzy. the yellow box of image E and image F processed by Wan [31] are gloomy and have lower contrast. Thus, our method has a good performance in contour enhancement and contrast improved.

E. QUANTITATIVE COMPARSION
In this part, we first analyze the evaluation metrics and then analyze the images in three different scenes.

1) QUANTITATIVE METRICS
In order to make a more comprehensive analysis and avoid the subjective impact of qualitative analysis, our method and the five comparison methods are analyzed through the three quantitative metrics: average gradient (AG), information entropy (IE) and enhancement by entropy (EME).
AG reflects the change rate of image gray scale and the brightness of image and represents the richness of image detail information. Typically, the larger the value of AG, the more detailed information the image contains. For an input image F(i, j), AG is defined as follows.
where ∇xF (i, j) and ∇yF (i, j) are the difference of F (i, j) along the x and y directions, M and N are the width and height of the input image, respectively.
IE represents the average amount of information in an image. For images with gray value between [0,255], the maximum information entropy is 8. Typically, the larger the value of IE, the more detailed information the image contains. It's defined as follows.
where IE is image information entropy, i is image gray value, and p(i) is the probability of image gray value i appearing in the whole image. EME represents the difference between the brightest and darkest gray values in the image and reflects the contrast of the image. Typically, the larger the value of EME, the more obvious the contrast of the image, indicating that the effect of image enhancement is more significant. Assuming that the image is divided into k 1 × k 2 modules, I w max;k,l and I w min;k,l are respectively the gray maximum and minimum of the central gray value co-ordinate (k, l), so EME is 2) EVALUATION RESULTS AND ANALYSIS According to the above formula, AG, IE and EME of image A-F are solved, and the results are shown in Table 2. From Table 2, it can be clearly seen that the BBHE [33] and HEEF [32] processed images have improved AG compared with the original image. The IE of the images processed by Wan [31] and HEEF [32] are lower than that of the original image. The IE of the images processed by other methods is higher than that of the original image and the IE of the images processed by our method is the largest, which indicates that HEEF [32] and Wan [31] will reduce the contrast of the image, other methods can improve the contrast of the image and the image processed by this method is higher. The EME of images processed by Wan [31] is lower than that of the original image, while the EME of images processed by other methods is improved, indicating that BBHE [33], DOTHE [29] and FCCE [30] can all improve the image quality, while Wan [31] reduces the image quality. The AG of the image A-F processed by our method is the largest than that of the original image and the other five algorithms in AG, IE and EME, indicating that the image processed by our method has the largest contrast, the most abundant detailed information and the best visual effect compared with other five methods. In addition, the running time is mainly used to measure the efficiency of the method in processing images. In Table 3, We give the running time of our method and five other methods in image A-F. In terms of running time, our method and Wan [31] consume more time while other five methods consume less time. This is mainly because both our method and Wan [31] need to use intelligent algorithms to process images, while other methods directly operate the gray value of images. Based on the above analysis, it can be concluded that this method can solve the problems of low contrast, fuzzy detail and poor image quality, but has longer running time than BBHE [33], HEEF [32], DOTHE [29] and FCCE [30].

IV. CONCLUSION
In this paper, an image enhancement method based on adaptive detail equalization is proposed. Our method consists of four parts: multi-scale convolution, adaptive bi-interval histogram equalization with details, adaptive limited Laplace operator and linear fusion, where multi-scale convolution and adaptive limited Laplace operator are mainly used to enhance image details, and adaptive bi-interval histogram equalization with details is mainly used to improve image brightness and contrast. The performance of our method is evaluated in three different scenes and compared with five other methods. The results show that our method has a good enhancement effect, improves the overall contrast of the image, enhances the details of the image, and improves the quality of the image. Especially, our method has a better performance on in both qualitative and quantitative than the other five methods. However, our method requires GA to obtain the optimal solution of threshold value, which results in a long running time and low efficiency of the method. In the next stage, we will focus on solving this problem.