WO2019205707A1 - 基于暗通道的线状自适应改进全局大气光的图像去雾方法 - Google Patents

基于暗通道的线状自适应改进全局大气光的图像去雾方法 Download PDF

Info

Publication number
WO2019205707A1
WO2019205707A1 PCT/CN2018/125153 CN2018125153W WO2019205707A1 WO 2019205707 A1 WO2019205707 A1 WO 2019205707A1 CN 2018125153 W CN2018125153 W CN 2018125153W WO 2019205707 A1 WO2019205707 A1 WO 2019205707A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
atmospheric light
dark
dark channel
channel
Prior art date
Application number
PCT/CN2018/125153
Other languages
English (en)
French (fr)
Inventor
黄鹤
宋京
郭璐
汪贵平
李昕芮
王会峰
许哲
崔博
黄莺
惠晓滨
Original Assignee
长安大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 长安大学 filed Critical 长安大学
Priority to US16/968,613 priority Critical patent/US11257194B2/en
Publication of WO2019205707A1 publication Critical patent/WO2019205707A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/73—Deblurring; Sharpening
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/80—Geometric correction
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/136—Segmentation; Edge detection involving thresholding
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/60—Analysis of geometric attributes
    • G06T7/66—Analysis of geometric attributes of image moments or centre of gravity
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/90—Determination of colour characteristics
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30181—Earth observation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30181—Earth observation
    • G06T2207/30192—Weather; Meteorology

Definitions

  • the invention belongs to the technical field of image processing, and in particular relates to an image de-fogging method based on a linear channel for improving global atmospheric light.
  • Images taken during smog weather will be affected by the weather and will degrade the quality. This is because dust particles and water vapor in the air will absorb and scatter light, causing the light intensity received by the sensor to change.
  • Scenery in foggy images is less clear than images taken on sunny days, which may result in limited image-based applications such as traffic safety monitoring and target recognition in aerial surveillance.
  • Image defogging technology can eliminate or reduce the impact of haze weather on image quality, so it has practical significance.
  • Image processing-based enhancement algorithms include histogram equalization and Retinex algorithm.
  • the physical model-based restoration algorithms include Tan algorithm, Fattal algorithm and He algorithm.
  • He et al. proposed a single-frame image dehazing algorithm based on the prior knowledge of dark primary colors at the CVPR conference, which achieved a good dehazing effect, and then improved the algorithm accordingly.
  • the algorithm applies the statistical principle of dark primary color to the physical model-atmospheric scattering model, and successively seeks the dark primary color map, atmospheric light value, rough propagation map and fine propagation map. Finally, it is substituted into the atmospheric scattering model to obtain the dehazing image.
  • the algorithm of He et al. when dealing with dense fog, may not be able to obtain the contrast between the dark region and the bright region after the image is processed due to uneven illumination.
  • the object of the present invention is to provide an image dehazing method based on a dark channel for linear adaptive improvement of global atmospheric light, which uses an atmospheric light map instead of the global atmospheric light value to solve the uneven distribution of atmospheric light in dense fog weather.
  • the problem of image distortion caused by the difference between the atmospheric fog and the near-fragment after defogging is too large, and the present invention can maintain the contrast of the dark area of the image and visualize the details of the scene in the bright area.
  • An image dehazing method based on dark channel-based linear adaptive improvement of global atmospheric light comprising the following steps:
  • Step 1 Obtain an image of haze under haze weather
  • Step 2 performing threshold segmentation on the haze image obtained in step 1 to obtain a binary image thereof;
  • Step 3 Calculate the image center of gravity (x 0 y 0 ) and the image center (0.5*h 0.5*w) for the binary image obtained in step 2, where h is the height of the binary image and w is the width of the binary image. Then, after dividing the horizontal and vertical coordinates of the center of gravity of the image and the center by h, w, respectively, the center of gravity (x 0 ' y 0 ') and the center (0.5 0.5) are obtained, and the slope k is defined:
  • Step 4 Find the dark channel of the smog image I obtained in step 1:
  • ⁇ (x, y) is the window of the neighborhood of the point (x, y)
  • I dark '(x, y) is the image of the dark channel
  • I C (x', y') is the image of the haze image I Monochrome channel image pixels
  • Step 5 Find the atmospheric light pattern A'(x, y) whose uniform change is obtained for the dark channel obtained in step 4;
  • Step 6 The atmospheric light map A'(x, y) obtained in step 5 is rotated counterclockwise according to the atmospheric light deflection degree ⁇ obtained in step 3 to obtain a final atmospheric light map A(x, y).
  • the light map is regularly distributed according to the direction of change of the fog;
  • Step 7 Obtain a defogged image from the atmospheric scattering model, where the atmospheric scattering model is as follows:
  • J is the clear image after defogging
  • t is the transmittance
  • A is the final atmospheric light map A(x, y) obtained in step 6.
  • ⁇ (x, y) is a 9 ⁇ 9 image block.
  • the atmospheric light map A'(x, y) whose uniform change is obtained in step 5 is as follows:
  • each dark channel picture I dark '(x, y) are sorted from large to small, and the first 0.1% minimum is taken as the atmospheric light value of the line, and each line of atmospheric light is sequentially obtained.
  • the initial atmospheric light map A 0 (x, y) is obtained; the initial atmospheric light map is filtered to obtain a uniformly changing atmospheric light map A'(x, y).
  • step 5 the initial atmospheric light pattern is filtered by means of mean filtering.
  • obtaining the defogging image in step 7 is specifically:
  • the dark primary color of the defogged image J approaches 0, namely:
  • J dark (x, y) is the dark channel pixel of the defogged image
  • ⁇ (x, y) is the window of the neighborhood of the point (x, y)
  • J C (x', y') is the foggy image J ( Monochrome channel image pixels of x, y);
  • the rough transmittance map is:
  • the factor ⁇ is 0.95.
  • the present invention has the following beneficial technical effects:
  • the invention adopts atmospheric light map to replace the traditional global atmospheric light value, and makes the atmospheric light value of the image linearly distributed.
  • the dark channel algorithm can solve the defogging problem of most foggy images, and the effect is good.
  • the brightness is too large and the distortion is serious, or
  • the near brightness value is too small and the details are lost.
  • the method uses an adaptive linear atmospheric light map that changes along the direction of the fog to replace the original global atmospheric light value, so that the dark channel dehazing algorithm can achieve good effects in a dense fog and a deep depth of field, and the defogging effect is good. It restores the vision better and has an ideal effect on image processing under smog weather. It has important significance for further processing of images and accurate image information acquisition.
  • Figure 1 is a schematic flow chart of the present invention
  • Figure 2 is a process diagram of the present invention in which (a) the original foggy image, (b) the binary image of the foggy image, the yellow center, the red center of gravity, and (c) the foggy image dark channel, (d ) the dark channel after rotation, (e) the atmospheric light pattern of the dark channel after rotation, (f) the atmospheric light pattern rotated back, (g) the atmospheric light pattern that changes uniformly after filtering, (h) the processed after using this method image;
  • Figure 3 is a comparison of the dehazing result of the present invention with a conventional dark channel, wherein (a) the original fogged image, (b) the conventional dark channel processed image, (c) the processed image of the present invention, (d) the conventional The dark channel portion is enlarged in detail, and (e) the details of the process of the present invention are enlarged.
  • an image re-fogging method for improving global atmospheric light based on linear adaptation of a dark channel includes the following steps:
  • Step 1 Obtain an image of the haze under smog weather.
  • Step 2 Step 1 obtains a haze image and performs threshold segmentation to obtain a binary image thereof;
  • the image is first converted into a grayscale image, and then the grayscale image is segmented into a binary image by Otsu algorithm threshold segmentation.
  • Step 3 Find the center of gravity (x 0 y 0 ) of the image and the center of the image (0.5*h 0.5*w) for the binary image obtained in step 2 (height h, width w). After the image center of gravity and the center corresponding coordinates are normalized by dividing by h, w, respectively, the center of gravity (x 0 ' y 0 ') and the center (0.5 0.5) are obtained, and the slope is defined:
  • the fog of the image is roughly distributed along the center line in a clockwise direction of ⁇ degree.
  • Step 4 After step 2, we perform gradation conversion on the image.
  • the binary image obtained by binary separation is divided into two parts: bright area and dark area. Then, through step 3, the center of gravity and center of the binary image are obtained. The center of gravity and the direction of the center line as the direction of change of atmospheric light to tilt the image
  • ⁇ (x, y) is the window of the neighborhood of the point (x, y), and I dark '(x, y) is the single image of the dark channel image I C (x', y') is the foggy image I Color channel image pixels;
  • the image dark channel I dark (x, y) is rotated counterclockwise according to the rotation angle ⁇ obtained in step 3 to obtain a dark channel I dark '(x, y) after the rotation.
  • Step 5 Calculate the atmospheric light distribution map of the dark channel obtained in step 4, as follows:
  • the initial atmospheric light map is filtered, and the mean filtering method is used to eliminate the abrupt change of each line of the atmospheric light map to obtain a uniform atmospheric light map A'(x, y).
  • Step 6 The uniform atmospheric light pattern A'(x, y) obtained in step 5 is rotated counterclockwise for the atmospheric light value deflection degree ⁇ obtained in step 3 to obtain a final atmospheric light map A(x, y). Atmospheric light patterns are regularly distributed according to the direction of the fog.
  • Step 7 The atmospheric scattering model commonly used in the study of defogging algorithms:
  • A is the final atmospheric light map A(x, y) obtained in step 6.
  • the dark primary color of the defogged image J should approach 0, namely:
  • J dark (x, y) is the dark channel pixel of the defogged image
  • ⁇ (x, y) is the window of the neighborhood of the point (x, y)
  • J C (x', y') is the foggy image J ( Monochrome channel image pixels of x, y);
  • the rough transmittance map is:
  • a fog-free clear image can be solved with I, t and A:
  • Figure 2 shows the process of processing a foggy image.
  • (a) is a foggy original image
  • (b) is a binary image of a foggy image, the line in the figure is the line connecting the center and the center of gravity
  • (c) is the dark channel of the foggy image
  • (d) After the counterclockwise rotation of the dark channel, the rotation angle is the line connecting the center of gravity and the center in the figure (b);
  • (e) is the atmospheric light pattern obtained by obtaining the atmospheric light value of each line on the dark channel after the rotation;
  • the atmospheric light pattern of Fig. (e) is rotated clockwise back and the atmospheric light pattern after cutting;
  • (g) is an atmospheric light pattern with uniform change after filtering;
  • (f) is a clear picture after processing.
  • FIG. 3(a) is the original foggy image, and the visible image has a distant view, which is blurred at a distance;
  • FIG. 3(b) adopts The results of the traditional dark channel theory treatment show that some fog is removed after the treatment, but some fog remains, especially the distant part makes the image blurred;
  • the processing result of the invention is shown in Fig. 3(c), the overall picture fog is basically clear and clear;
  • 3(d) is a magnified view of the foreground of the traditional dark channel. It can be seen that there is still some fog in the picture to make the picture more blurred.
  • Figure (e) shows the details of the foreground image of the image after processing in the present invention. Compared with Figure 3(d), it can be found that the mist removal is more thorough and the definition is improved.
  • Table 1 is a comparison table of fogging image processing effect parameters by different algorithms. It can be seen from Table 1 that the image is further modified by the dark channel dehazing method after the method of defogging, ambiguity, average gradient, contrast, information entropy All of them are improved. It can be seen that the effect of the fog image processing is further improved, and the result is good, which is superior to the traditional dark channel defogging method, and has important significance for further research on image defogging and fog image information extraction.
  • the image dehazing algorithm of the improved dark channel of the invention has good defogging effect, better restoration of the foreground, ideal image processing under the smog weather, further processing of the image and accurate acquisition of the image information. Significance.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Geometry (AREA)
  • Image Processing (AREA)

Abstract

一种基于暗通道的线状自适应改进全局大气光的图像去雾方法，首先获取雾霾天气下的雾霾图像，然后通过对图像的二值图像求取重心中心连线斜率的方法获得图像大气光变化角度θ，再求取沿大气光变化方向θ规律变化的线形大气光图，之后再通过大气散射模型求解无雾图像，输出处理后的雾霾天气下的有雾图像。该方法既满足了在浓雾或者景深较深情况下远处景物不失真，又保留了近处景物的细节，对后续的雾霾图像处理及信息提取尤为重要。

Description

基于暗通道的线状自适应改进全局大气光的图像去雾方法 技术领域
本发明属于图像处理技术领域，具体涉及一种基于暗通道的线状自适应改进全局大气光的图像去雾方法。
背景技术
雾霾天气时拍摄的图像，会受到天气的影响而降低质量，这是因为雾霾天气时空气中灰尘颗粒、水汽会吸收和散射光，导致传感器接收到的光强发生变化。有雾图像中景物的清晰程度要比晴天拍摄的图像低，可能会导致一些基于图像的应用受到限制，如交通安全监控、航拍监视中的目标识别等。图像去雾技术能够消除或减少雾霾天气对图像质量的影响，因此具有实际意义。
目前，目前已经有一些算法能够实现单幅图像去雾，这些算法在原理上可以分为基于图像处理的增强方法和基于物理模型的复原方法。基于图像处理的增强算法有直方图均衡化和Retinex算法等，基于物理模型的复原算法有Tan算法、Fattal算法和He算法等。
2009年，He等人在CVPR会议上提出了基于暗原色先验知识的单帧图像去雾算法，取得了较好的去雾效果，后又对算法进行了相应改进。该算法将暗原色先验统计规律应用在物理模型——大气散射模型上，先后求暗原色图、大气光值、粗略传播图和精细传播图，最后代入大气散射模型求得去雾图像。但是经过大量实验发现，he等人的算法在处理浓雾的情况时，会由于光照不均导致选取的大气光值在处理完图像后暗区对比度和亮区的细节不能兼得。
发明内容
本发明的目的在于提供一种基于暗通道的线状自适应改进全局大气光的图像去雾方法，使用大气光图代替全局大气光值，以解决在浓雾天气时大气 光分布不均而导致去雾后浓雾与近景部分因大气光值相差太大造成的图像失真问题，本发明既能保持图像暗区的对比度，又能使亮区的景物细节得以显现。
为达到上述目的，本发明采用如下技术方案：
基于暗通道的线状自适应改进全局大气光的图像去雾方法，包括以下步骤：
步骤1：获取雾霾天气下雾霾图像；
步骤2：对步骤1得到的雾霾图像进行阈值分割求取其二值图像；
步骤3：对步骤2得到的二值图像求取其图像重心(x 0 y 0)与图像中心(0.5*h 0.5*w)，其中h为二值图像的高，w为二值图像的宽，然后对图像重心和中心对应的横坐标和纵坐标分别除以h，w进行归一化后，得到重心(x 0′ y 0′)和中心(0.5 0.5)，定义斜率k：
Figure PCTCN2018125153-appb-000001
定义θ为大气光值偏转度：
θ＝arctan(1/k)
步骤4：对步骤1得到的雾霾图像I求取其暗通道：
I dark(x,y)＝min C∈{r,g,b}(min (x′,y′)∈Ω(x,y)(I C(x′,y′)))
式中：Ω(x,y)为点(x,y)邻域的窗口，I dark′(x,y)为暗通道图像像，I C(x′,y′)为雾霾图像I的单色通道图像像素；
将图像暗通道I dark(x,y)按照步骤3得到的旋转角θ逆时针旋转，得到旋转后的暗通道I dark′(x,y)；
步骤5：对步骤4得到的暗通道求取其变化均匀的大气光图A′(x,y)；
步骤6：对步骤5得到的变化均匀的大气光图A′(x,y)按照步骤3得到的 大气光值偏转度θ逆时针旋转，得到最终大气光图A(x,y)，此大气光图按照雾的浓淡变化方向规律分布；
步骤7：由大气散射模型求取去雾图像，其中大气散射模型如下：
I(x,y)＝J(x,y)t(x,y)+A(1-t(x,y))
其中J为去雾后清晰图像，t是透射率，A即为步骤6求得的最终大气光图A(x,y)。
进一步地，步骤4中Ω(x,y)为9×9的图像块。
进一步地，步骤5中求取其变化均匀的大气光图A′(x,y)具体如下：
将旋转后暗通道图片I dark′(x,y)的每行暗通道值从大到小进行排序，取其前0.1％的最小值作为该行的大气光值，依次求取每行大气光值，得到初始大气光图A 0(x,y)；对初始大气光图进行滤波，得到变化均匀的大气光图A′(x,y)。
进一步地，步骤5中采用均值滤波的方法对初始大气光图进行滤波。
进一步地，步骤7中求取去雾图像具体为：
根据暗原色统计规律，去雾图像J的暗原色趋近于0，即：
J dark(x,y)＝min(min (x′,y′)∈Ω(x,y)(J C(x′,y′)))＝0
其中J dark(x,y)为去雾图像的暗通道像素，Ω(x,y)为点(x,y)邻域的窗口，J C(x′,y′)为有雾图像J(x,y)的单色通道图像像素；
而A恒为正数，则有：
min C(min (x′,y′)∈Ω(x,y)(J C(x′,y′)/A))＝0
得粗略透射率图为：
t′(x,y)＝1-min C(min (x′,y′)∈Ω(x,y)(I C(x′,y′)/A))
晴朗天气下，远方的景物也会有少许雾气遮罩，为了使去雾效果不失真，加入因子ω：
t(x,y)＝1-ωmin C(min (x′,y′)∈Ω(x,y)(I C(x′,y′)))
采用I，t和A解出无雾清晰图像J：
J(x,y)＝(I(x,y)-A(x,y))/t(x,y)+A(x,y)
并输出无雾清晰图像J。
进一步地，因子ω取0.95。
与现有技术相比，本发明具有以下有益的技术效果：
本发明采用大气光图来代替传统的全局大气光值，使图像的大气光值线状分布，在传统去雾算法中，暗通道算法可以解决大多数有雾图像的去雾问题，且效果良好，但对景深较深，远处景物雾浓度远大于近处的图像进行去雾处理时，由于全局使用同一个大气光至使图像近处效果良好时，远处亮度过大而失真严重，或者远处效果良好时近处亮度值太小而导致细节丢失。本方法使用自适应的沿雾浓淡方向变化的线形大气光图替代原本全局大气光值，使暗通道去雾算法在浓雾及景深较大的地方也能取得良好的效果，去雾效果好，对远景复原较好，对雾霾天气下图片处理有理想效果，对图像的进一步处理以及准确获取图像信息有着重要的意义。
附图说明
图1是本发明的流程示意图；
图2是本发明中各个过程图，其中，(a)原始有雾图像，(b)有雾图像的二值图像，黄色为中心，红色为重心，(c)有雾图像暗通道，(d)旋转后的暗通道，(e)旋转后暗通道的大气光图，(f)旋转回来的大气光图，(g)滤波后变化均匀的大气光图，(h)使用本方法处理后的图像；
图3是本发明的去雾结果与传统暗通道的对比图，其中，(a)原始有雾图像，(b)传统暗通道处理后图像，(c)本发明处理后图像，(d)传统暗通道部分细节放大，(e)本发明处理后部分细节放大。
具体实施方式
下面结合附图对本发明作进一步详细描述：
参见图1，基于暗通道的线状自适应改进全局大气光的图像去雾方法，包括以下步骤：
步骤1：获取雾霾天气下雾霾图像。
利用图像采集设备，获取雾霾天气下降质的雾霾图像。
步骤2：将步骤1得到雾霾图像进行阈值分割求取其二值图像；
先将图片转换为灰度图像，再将灰度图像用Otsu算法阈值分割转换为二值图像。
步骤3：对步骤2得到的二值图像(高为h，宽为w)求取其图像的重心(x 0 y 0)与图像中心(0.5*h 0.5*w)。对图像重心和中心对应坐标分别除以h，w进行归一化后，得到重心(x 0′ y 0′)和中心(0.5 0.5)，定义斜率：
Figure PCTCN2018125153-appb-000002
定义θ为大气光值偏转度：
θ＝arctan(1/k)
此时图像中雾的浓淡则大致沿中线顺时针偏转θ度方向变化分布。
步骤4：经过步骤2我们对图像进行灰度转换，二值分割后得到的二值图像即分为亮区和暗区两部分，再通过步骤3，求取二值图像的重心与中心，采用重心与中心连线方向作为大气光变化方向来使图像倾斜
对步骤1得到的雾霾图像求取其暗通道
I dark(x,y)＝min C∈{r,g,b}(min (x′,y′)∈Ω(x,y)(I C(x′,y′)))
式中：Ω(x,y)为点(x,y)邻域的窗口，I dark′(x,y)为暗通道图像像I C(x′,y′)为有雾图像I的单色通道图像像素；
将图像暗通道I dark(x,y)按照步骤3得到的旋转角θ逆时针旋转，得到旋转后的暗通道I dark′(x,y)。
步骤5：对步骤4得到的暗通道求取其大气光分布图，具体如下：
将旋转后暗通道图片I dark′(x,y)的每行暗通道值从大到小进行排序，取其前百分之0.1里的最小值作为该行的大气光值，依次求取每行大气光值，得到初始大气光图A 0(x,y)。
对初始大气光图进行滤波，采用均值滤波的方法使大气光图消除每行突变，得到变化均匀的大气光图A′(x,y)。
步骤6：对步骤5得到的变化均匀的大气光图A′(x,y)按照步骤3得到的为大气光值偏转度θ逆时针旋转，得到最终大气光图A(x,y)，此大气光图按照雾的浓淡变化方向规律分布。
步骤7：由去雾算法研究中常用的大气散射模型：
I(x,y)＝J(x,y)t(x,y)+A(1-t(x,y))
其中J为去雾后清晰图像，t是透射率。A即为步骤6求得的最终大气光图A(x,y)。
根据暗原色统计规律，去雾图像J的暗原色应该趋近于0，即：
J dark(x,y)＝min C(min (x′,y′)∈Ω(x,y)(J C(x′,y′)))＝0
其中J dark(x,y)为去雾图像的暗通道像素，Ω(x,y)为点(x,y)邻域的窗口，J C(x′,y′)为有雾图像J(x,y)的单色通道图像像素；
而A恒为正数，则有：
min C(min (x′,y′)∈Ω(x,y)(J C(x′,y′)/A))＝0
可得粗略透射率图为：
t′(x,y)＝1-min C(min (x′,y′)∈Ω(x,y)(I C(x′,y′)/A))
晴朗天气下，远方的景物也会有少许雾气遮罩，为了使去雾效果不失真，还需在上式中加入一个因子ω，一般将ω取为0.95左右。
t(x,y)＝1-ωmin C(min (x′,y′)∈Ω(x,y)(I C(x′,y′)))
可以用I，t和A解出无雾清晰图像J：
J(x,y)＝(I(x,y)-A(x,y))/t(x,y)+A(x,y)
输出无雾清晰图像J。
图2即为本算法处理有雾图像的过程。其中(a)为有雾原图；(b)为有雾图像的二值图像，图中的连线即为中心与重心的连线；(c)为有雾图像的暗通道；(d)为暗通道逆时针旋转后，旋转角度为图(b)中重心与中心连线；(e)为对旋转后的暗通道求取每行的大气光值得到的大气光图；(f)为将图(e)的大气光图顺时针旋转回并裁剪后的大气光图；(g)为滤波后变化均匀的大气光图；(f)为处理后的清晰图片。
由图3的处理结果对比和局部放大图可以更直观地看到处理效果，图3(a)为原始有雾图像，可见图像中有远景近景，较远处一片模糊；图3(b)采用传统暗通道理论处理结果，可见处理后消除部分雾气，但仍保留有部分雾气，尤其远景部分使图像模糊；采用本发明处理结果如图3(c)，整体图片雾气基本清楚，较为清晰；图3(d)为传统暗通道的远景细节放大，可见图中仍有部分雾气使图片较为模糊。图(e)为本发明处理后图像远景细节放大，与图3(d)相比较可发现雾气去除更为彻底，清晰度提高。
表1为不同算法对有雾的图像处理效果参数对比表，由表1可以看出，图像再经本发明改进后的暗通道去雾方法去雾后，模糊度、平均梯度、对比度、信息熵都有所提高，可见本发明对有雾图像处理的效果进一步改进，结 果良好，已优于传统的暗通道去雾方法，对进一步研究图像去雾，有雾图像信息提取等方面有着重要意义。
表1不同算法对有雾的图像处理效果参数对比表
Figure PCTCN2018125153-appb-000003
综上所述，本发明的改进暗通道的图像去雾算法，去雾效果好，对远景复原较好，对雾霾天气下图片处理有理想效果，对图像的进一步处理以及准确获取图像信息有着重要的意义。

Claims (6)

  1. 基于暗通道的线状自适应改进全局大气光的图像去雾方法，其特征在于，包括以下步骤：
    步骤1：获取雾霾天气下雾霾图像；
    步骤2：对步骤1得到的雾霾图像进行阈值分割求取其二值图像；
    步骤3：对步骤2得到的二值图像求取其图像重心(x 0 y 0)与图像中心(0.5*h 0.5*w)，其中h为二值图像的高，w为二值图像的宽，然后对图像重心和中心对应的横坐标和纵坐标分别除以h，w进行归一化后，得到重心(x 0′ y 0′)和中心(0.5 0.5)，定义斜率k：
    Figure PCTCN2018125153-appb-100001
    定义θ为大气光值偏转度：
    θ＝arctan(1/k)
    步骤4：对步骤1得到的雾霾图像I求取其暗通道：
    I dark(x,y)＝min C∈{r,g,b}(min (x′,y′)∈Ω(x,y)(I C(x′,y′)))
    式中：Ω(x,y)为点(x,y)邻域的窗口，I dark′(x,y)为暗通道图像像，I C(x′,y′)为雾霾图像I的单色通道图像像素；
    将图像暗通道I dark(x,y)按照步骤3得到的旋转角θ逆时针旋转，得到旋转后的暗通道I dark′(x,y)；
    步骤5：对步骤4得到的暗通道求取其变化均匀的大气光图A′(x,y)；
    步骤6：对步骤5得到的变化均匀的大气光图A′(x,y)按照步骤3得到的大气光值偏转度θ逆时针旋转，得到最终大气光图A(x,y)，此大气光图按照雾的浓淡变化方向规律分布；
    步骤7：由大气散射模型求取去雾图像，其中大气散射模型如下：
    I(x,y)＝J(x,y)t(x,y)+A(1-t(x,y))
    其中J为去雾后清晰图像，t是透射率，A即为步骤6求得的最终大气光图A(x,y)。
  2. 根据权利要求1所述的基于暗通道的线状自适应改进全局大气光的图像去雾方法，其特征在于，步骤4中Ω(x,y)为9×9的图像块。
  3. 根据权利要求1所述的基于暗通道的线状自适应改进全局大气光的图像去雾方法，其特征在于，步骤5中求取其变化均匀的大气光图A′(x,y)具体如下：
    将旋转后暗通道图片I dark′(x,y)的每行暗通道值从大到小进行排序，取其前0.1％的最小值作为该行的大气光值，依次求取每行大气光值，得到初始大气光图A 0(x,y)；对初始大气光图进行滤波，得到变化均匀的大气光图A′(x,y)。
  4. 根据权利要求3所述的基于暗通道的线状自适应改进全局大气光的图像去雾方法，其特征在于，步骤5中采用均值滤波的方法对初始大气光图进行滤波。
  5. 根据权利要求1所述的基于暗通道的线状自适应改进全局大气光的图像去雾方法，其特征在于，步骤7中求取去雾图像具体为：
    根据暗原色统计规律，去雾图像J的暗原色趋近于0，即：
    J dark(x,y)＝min(min (x′,y′)∈Ω(x,y)(J C(x′,y′)))＝0
    其中J dark(x,y)为去雾图像的暗通道像素，Ω(x,y)为点(x,y)邻域的窗口，J C(x′,y′)为有雾图像J(x,y)的单色通道图像像素；
    而A恒为正数，则有：
    min C(min (x′,y′)∈Ω(x,y)(J C(x′,y′)/A))＝0
    得粗略透射率图为：
    t′(x,y)＝1-min C(min (x′,y′)∈Ω(x,y)(I C(x′,y′)/A))
    晴朗天气下，远方的景物也会有少许雾气遮罩，为了使去雾效果不失真，加入因子ω：
    t(x,y)＝1-ωmin C(min (x′,y′)∈Ω(x,y)(I C(x′,y′)))
    采用I，t和A解出无雾清晰图像J：
    J(x,y)＝(I(x,y)-A(x,y))/t(x,y)+A(x,y)
    并输出无雾清晰图像J。
  6. 根据权利要求5所述的基于暗通道的线状自适应改进全局大气光的图像去雾方法，其特征在于，因子ω取0.95。
PCT/CN2018/125153 2018-04-26 2018-12-29 基于暗通道的线状自适应改进全局大气光的图像去雾方法 WO2019205707A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US16/968,613 US11257194B2 (en) 2018-04-26 2018-12-29 Method for image dehazing based on adaptively improved linear global atmospheric light of dark channel

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201810387076.2A CN108765309B (zh) 2018-04-26 2018-04-26 基于暗通道的线状自适应改进全局大气光的图像去雾方法
CN201810387076.2 2018-04-26

Publications (1)

Publication Number Publication Date
WO2019205707A1 true WO2019205707A1 (zh) 2019-10-31

Family

ID=64011904

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/125153 WO2019205707A1 (zh) 2018-04-26 2018-12-29 基于暗通道的线状自适应改进全局大气光的图像去雾方法

Country Status (3)

Country Link
US (1) US11257194B2 (zh)
CN (1) CN108765309B (zh)
WO (1) WO2019205707A1 (zh)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111325688A (zh) * 2020-02-18 2020-06-23 长安大学 融合形态学聚类优化大气光的无人机图像去雾方法
CN111539891A (zh) * 2020-04-27 2020-08-14 高小翎 单张遥感图像的波段自适应除雾优化处理方法
CN112927157A (zh) * 2021-03-08 2021-06-08 电子科技大学 采用加权最小二乘滤波的改进暗通道去雾方法
CN113012062A (zh) * 2021-03-04 2021-06-22 西安电子科技大学 一种高清视频实时快速去雾方法
CN113628131A (zh) * 2021-07-22 2021-11-09 济南驰昊电力科技有限公司 雾天场景下变电站指针式油位表的智能识别方法
CN115861133A (zh) * 2023-02-22 2023-03-28 山东晋工科技有限公司 一种钻凿劈裂一体机的远程遥控无人驾驶系统
CN116703787A (zh) * 2023-08-09 2023-09-05 中铁建工集团第二建设有限公司 一种建筑施工安全风险预警方法及系统

Families Citing this family (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108765309B (zh) * 2018-04-26 2022-05-17 西安汇智信息科技有限公司 基于暗通道的线状自适应改进全局大气光的图像去雾方法
CN109961413B (zh) * 2019-03-21 2023-05-12 平顶山学院 大气光方向优化估计的图像去雾迭代算法
CN112419163B (zh) * 2019-08-21 2023-06-30 中国人民解放军火箭军工程大学 一种基于先验知识和深度学习的单张图像弱监督去雾方法
CN113139922B (zh) * 2021-05-31 2022-08-02 中国科学院长春光学精密机械与物理研究所 图像去雾方法及去雾装置
US11803942B2 (en) 2021-11-19 2023-10-31 Stmicroelectronics (Research & Development) Limited Blended gray image enhancement
CN115063404B (zh) * 2022-07-27 2022-11-08 建首(山东)钢材加工有限公司 基于x射线探伤的耐候钢焊缝质量检测方法
CN115409740A (zh) * 2022-11-01 2022-11-29 国网湖北省电力有限公司 一种基于暗通道先验引导图像滤波去除图像雾霾的方法
CN115456913A (zh) * 2022-11-07 2022-12-09 四川大学 一种夜间雾图去雾方法及装置
CN115456915B (zh) * 2022-11-10 2023-05-09 深圳深知未来智能有限公司 基于3DLut的图像去雾处理方法、系统及可存储介质
CN115937144B (zh) * 2022-12-08 2023-08-25 郑州大学 一种胸腔镜术中图像处理方法及系统
CN116596805B (zh) * 2023-07-14 2023-09-29 山东大学 一种基于场景物体与大气光偏振态差异的偏振去雾方法
CN116630349B (zh) * 2023-07-25 2023-10-20 山东爱福地生物股份有限公司 基于高分辨率遥感图像的秸秆还田区域快速分割方法
CN116739608B (zh) * 2023-08-16 2023-12-26 湖南三湘银行股份有限公司 基于人脸识别方式的银行用户身份验证方法及系统
CN116977327B (zh) * 2023-09-14 2023-12-15 山东拓新电气有限公司 一种滚筒驱动带式输送机烟雾检测方法及系统

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160005152A1 (en) * 2014-07-01 2016-01-07 Adobe Systems Incorporated Multi-Feature Image Haze Removal
CN105654440A (zh) * 2015-12-30 2016-06-08 首都师范大学 基于回归模型的快速单幅图像去雾算法及系统
CN106548461A (zh) * 2016-10-25 2017-03-29 湘潭大学 图像去雾方法
CN107451962A (zh) * 2017-07-03 2017-12-08 山东财经大学 一种图像去雾方法及装置
CN108765309A (zh) * 2018-04-26 2018-11-06 长安大学 基于暗通道的线状自适应改进全局大气光的图像去雾方法

Family Cites Families (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8970691B2 (en) * 2011-08-26 2015-03-03 Microsoft Technology Licensing, Llc Removal of rayleigh scattering from images
US8885962B1 (en) * 2012-07-23 2014-11-11 Lockheed Martin Corporation Realtime long range imaging scatter reduction
US20170178297A1 (en) * 2014-02-19 2017-06-22 Yissum Research Development Company Of The Hebrew University Of Jerusalem Ltd. Method and system for dehazing natural images using color-lines
US9361670B2 (en) * 2014-09-04 2016-06-07 National Taipei University Of Technology Method and system for image haze removal based on hybrid dark channel prior
US9288458B1 (en) * 2015-01-31 2016-03-15 Hrl Laboratories, Llc Fast digital image de-hazing methods for real-time video processing
US20180122051A1 (en) * 2015-03-30 2018-05-03 Agency For Science, Technology And Research Method and device for image haze removal
CN105976330B (zh) * 2016-04-27 2019-04-09 大连理工大学 一种嵌入式雾天实时视频稳像方法
CN106373133B (zh) * 2016-08-31 2019-02-26 重庆广播电视大学 一种基于暗通道去雾算法的农田插秧检测方法及其系统
US10367976B2 (en) * 2017-09-21 2019-07-30 The United States Of America As Represented By The Secretary Of The Navy Single image haze removal
CN107767354B (zh) * 2017-12-08 2020-07-07 福州大学 一种基于暗原色先验的图像去雾算法
TWI674804B (zh) * 2018-03-15 2019-10-11 國立交通大學 視訊除霧處理裝置及方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160005152A1 (en) * 2014-07-01 2016-01-07 Adobe Systems Incorporated Multi-Feature Image Haze Removal
CN105654440A (zh) * 2015-12-30 2016-06-08 首都师范大学 基于回归模型的快速单幅图像去雾算法及系统
CN106548461A (zh) * 2016-10-25 2017-03-29 湘潭大学 图像去雾方法
CN107451962A (zh) * 2017-07-03 2017-12-08 山东财经大学 一种图像去雾方法及装置
CN108765309A (zh) * 2018-04-26 2018-11-06 长安大学 基于暗通道的线状自适应改进全局大气光的图像去雾方法

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111325688A (zh) * 2020-02-18 2020-06-23 长安大学 融合形态学聚类优化大气光的无人机图像去雾方法
CN111325688B (zh) * 2020-02-18 2023-05-05 西安汇智信息科技有限公司 融合形态学聚类优化大气光的无人机图像去雾方法
CN111539891A (zh) * 2020-04-27 2020-08-14 高小翎 单张遥感图像的波段自适应除雾优化处理方法
CN113012062A (zh) * 2021-03-04 2021-06-22 西安电子科技大学 一种高清视频实时快速去雾方法
CN112927157A (zh) * 2021-03-08 2021-06-08 电子科技大学 采用加权最小二乘滤波的改进暗通道去雾方法
CN112927157B (zh) * 2021-03-08 2023-08-15 电子科技大学 采用加权最小二乘滤波的改进暗通道去雾方法
CN113628131A (zh) * 2021-07-22 2021-11-09 济南驰昊电力科技有限公司 雾天场景下变电站指针式油位表的智能识别方法
CN115861133A (zh) * 2023-02-22 2023-03-28 山东晋工科技有限公司 一种钻凿劈裂一体机的远程遥控无人驾驶系统
CN116703787A (zh) * 2023-08-09 2023-09-05 中铁建工集团第二建设有限公司 一种建筑施工安全风险预警方法及系统
CN116703787B (zh) * 2023-08-09 2023-10-31 中铁建工集团第二建设有限公司 一种建筑施工安全风险预警方法及系统

Also Published As

Publication number Publication date
US20210049744A1 (en) 2021-02-18
US11257194B2 (en) 2022-02-22
CN108765309B (zh) 2022-05-17
CN108765309A (zh) 2018-11-06

Similar Documents

Publication Publication Date Title
WO2019205707A1 (zh) 基于暗通道的线状自适应改进全局大气光的图像去雾方法
CN106530237B (zh) 一种图像增强方法
CN106846263B (zh) 基于融合通道且对天空免疫的图像去雾方法
CN106157267B (zh) 一种基于暗通道先验的图像去雾透射率优化方法
CN109255759B (zh) 基于天空分割和透射率自适应修正的图像去雾方法
KR102104403B1 (ko) 단일영상 내의 안개 제거 방법 및 장치
CN107301623B (zh) 一种基于暗通道和图像分割的交通图像去雾方法及系统
Gao et al. Sand-dust image restoration based on reversing the blue channel prior
CN108389175B (zh) 融合变差函数和颜色衰减先验的图像去雾方法
CN107358585B (zh) 基于分数阶微分及暗原色先验的雾天图像增强方法
CN109087254B (zh) 无人机航拍图像雾霾天空和白色区域自适应处理方法
Pei et al. Effective image haze removal using dark channel prior and post-processing
CN111145105B (zh) 一种图像快速去雾方法、装置、终端及存储介质
CN110211067A (zh) 一种用于uuv近海面可见光图像去雾方法
CN111161167B (zh) 基于中通道补偿和自适应大气光估计的单幅图像去雾方法
CN111598814B (zh) 基于极端散射通道的单图像去雾方法
CN112053298B (zh) 一种图像去雾方法
CN105023246B (zh) 一种基于对比度和结构相似度的图像增强方法
CN111598800A (zh) 基于空间域同态滤波和暗通道先验的单幅图像去雾方法
CN111325688A (zh) 融合形态学聚类优化大气光的无人机图像去雾方法
CN108765316B (zh) 雾气浓度自适应判断方法
CN108805826B (zh) 改善去雾效果的方法
Wang et al. Single-image dehazing using color attenuation prior based on haze-lines
CN110349113B (zh) 一种基于暗原色先验改进的自适应图像去雾方法
CN107203979B (zh) 一种低照度图像增强的方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18915909

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 18915909

Country of ref document: EP

Kind code of ref document: A1