CN105513025A - Improved rapid demisting method - Google Patents

Abstract

The invention discloses an improved fast defogging method, which is divided into three parts: the first part is to carry out mean value down-sampling processing on YUV data, and convert it into a small image in RGB format; the second part is to solve the problem caused by down-sampling For the problem of sudden blackening of the edge, the maximum value filtering is performed on the transmittance of the small image; the third part is aimed at the block effect caused by downsampling, and bilinear interpolation upsampling is used to avoid block effect. The method proposed by the invention can achieve real-time processing on the PC for the video in YUV format with 500W pixels. The present invention proposes an effective improvement method for the side effects of black borders caused by down-sampling and up-sampling, which can effectively avoid such side effects. The invention proposes bilinear interpolation and up-sampling to remove the block effect caused by down-sampling.

Description

An Improved Fast Dehazing Method

technical field

The invention relates to the technical fields of digital image processing and intelligent transportation, in particular to an improved fast defogging method.

Background technique

In recent years, environmental pollution has been very serious, especially more and more smoggy weather. Fog and haze are particles suspended in the air, whose scattering effect greatly reduces the image quality. The reduction of image quality directly affects the functions of outdoor image acquisition systems such as speedometers and driving recorders. It not only reduces the visibility of the image, but also causes difficulties in the subsequent image processing algorithms (such as license plate recognition, feature extraction, image analysis, etc.). Nowadays, most of the image acquisition devices are high-definition cameras, and the collected images are clear, but the data volume of the images is also large, and it takes a long time to process them. Therefore it is very important to invent a method that can quickly remove the fog.

However, the performance and function of the existing dehazing algorithm are a pair of contradictions. The algorithm with good effect has high complexity and is difficult to achieve real-time performance; the algorithm with fast speed is relatively simple, but the processing result is not ideal, which often causes color distortion, etc. side effect. Therefore, it is necessary to make a trade-off between performance and function, and propose a method that has good performance in terms of performance and function.

There are many existing image defogging methods, which can be generally divided into two categories: methods based on image enhancement and methods based on physical models. The method of image enhancement is to enhance the degraded image to improve the quality of the image. This method is relatively simple and the processing speed is fast, but the processing effect is not ideal, which may cause the loss of part of the image information, resulting in image distortion. Based on the physical model method, this method establishes an atmospheric scattering model by studying the scattering effect of atmospheric suspended particles on light, understands the physical mechanism of image degradation, and inverts and restores a fog-free image. Nowadays, many dehazing algorithms are based on physical model proposed. Prior Art 1 (HeK, SunJ, TangX. Singleimagehazeremovingdarkchannelprior.IEEETransactionsonPatternAnalysisandMachineIntelligence, 2011, 33(12): 2341-2353) found a priori law named dark channel priori by observing the statistical characteristics of a large number of haze-free images. This method has a very good performance in the processing effect and opens up a new field of image dehazing. However, soft matting is used to refine the transmittance map in this paper, which is very complex and time-consuming. Later, the author of the article uses guided filtering instead of soft matting. The dehazing effect is equivalent, but the processing speed is increased by about 100 times. But even if guided filtering is used to dehaze high-definition video, there is still a big gap in real-time processing. In prior art 2 (TarelJP, HautiereN.Fastvisibilityrestorationfromasinglecolororgraylevelimage.In:Proceedingofthe12thIEEEInternationalConferenceonComputerVision, 2009.Kyoto:IEEE, 2009.2201-2208), a method for fast dehazing is proposed, using double-median filtering instead of the minimum Value filtering and guided filtering greatly simplify the processing process and improve efficiency. However, the median filter is not a good edge-preserving filter algorithm, and a sudden change in the depth of field in a local area will produce a halo effect. Moreover, there are many parameters in the algorithm, which cannot realize self-adaptive adjustment, and requires manual testing and adjustment, which is limited in practical application.

There are also many research institutions or universities in China that have achieved good results in fog removal. Prior art three (DSP-based fog video processing system and method CN103347171A) designed a set of processing system for fog video, including the design of hardware and software, in the program optimization, some parameters that change slowly are updated every other time. method to reduce processing time. However, the author finally only said that the simulation processing for 432*283 images takes 3.622s, and did not clearly indicate whether the high-definition video can achieve real-time performance on the DSP platform. In prior art 4 (CN104240192A, title of invention: a fast single image dehazing algorithm), the dehazing model is quickly synthesized under specific conditions by using the minimum value, image gradient and dark channel map in the RGB three channels of the image color space The required transmission map replaces the step of solving the transmission map by the soft matting method in the original dark channel dehazing algorithm, and optimizes the calculation of the dark channel. This method changes the operation of the original large-scale sparse matrix into the comparison of the corresponding pixels of several different information images, the amount of calculation is greatly reduced, and in most cases, the same ideal results as the original algorithm can be obtained.

Contents of the invention

The purpose of the present invention is: 1) for the traffic video of 500W pixel YUV format, the method that the present invention proposes can reach real-time processing on PC; This side effect can be well avoided. 3) For the blocking effect caused by downsampling, the present invention uses bilinear interpolation to upsampling, which can avoid blocking effect.

The technical scheme adopted in the present invention is: an improved quick defogging method, the steps of which are as follows:

Step 1), input the image of a frame YUV format;

Step 2), YUV data mean value downsampling, and convert to RGB format;

Perform mean downsampling processing on the YUV data to obtain an image in YUV format after downsampling, and convert the image in YUV format into an image in RGB format;

Step 3), calculate the transmittance of the small image based on the dark channel prior principle;

For the small RGB image in step 2), obtain the dark channel through minimum value filtering, estimate the atmospheric light value, calculate the estimated transmittance, and finally perform guided filtering on the estimated transmittance to obtain a refined transmittance;

Step 4), maximum value filtering;

Perform bilinear interpolation up-sampling after the maximum value filtering process with a radius of 1 on the transmittance of the small image to avoid the phenomenon of black edges;

Step 5), bilinear interpolation upsampling;

Using bilinear interpolation upsampling, linear interpolation is performed between several points, so that the image after upsampling changes smoothly and the block effect can be eliminated;

Step 6) directly restores to a haze-free YUV image;

Use the large transmittance image after interpolation in step 5) to reconstruct the defogged image, and directly use the YUV format for reconstruction, omitting the time and space of intermediate YUV and RGB conversion;

Step 7) Input the next frame of YUV image.

The advantages and positive effects of the technical solution of the present invention are:

1) YUV mean downsampling and conversion to RGB format;

The mean value downsampling process is performed on the YUV data. The mean value downsampling can guarantee the information of the original image to a certain extent, obtain the image in YUV format after downsampling, and convert the image in YUV format into an image in RGB format. This can greatly reduce the time consumption caused by calculation and save a part of storage space.

2) Solve the problem of black borders on sudden changes;

The use of downsampling and upsampling will bring certain side effects, especially for billboards or vehicles with darker colors in large sky areas, which will cause excessive defogging at the junction of billboards or car bodies with the sky after processing situation, that is, the appearance of black borders. In the present invention, the maximum value filtering process with a radius of 1 is performed on the transmittance in the algorithm, which can effectively avoid the black edge phenomenon.

3) Avoid the block effect of downsampling;

In the present invention, bilinear interpolation up-sampling is performed on the small transmittance image, and the obtained large transmittance image is used to reconstruct the image, so that a good defogging effect can be obtained finally, and there is no such side effect as block effect.

Description of drawings

Fig. 1 is a schematic diagram of bilinear interpolation upsampling;

Fig. 2 is the flowchart of improved method;

Figure 3 shows the results of dehazing.

detailed description

The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

The technical solution of the present invention is divided into three parts: the first part is to carry out the average value down-sampling process to the YUV data, and convert it into a small image in RGB format; The transmittance of the small image is filtered by the maximum value; the third part is to avoid the blocking effect caused by downsampling by using bilinear interpolation and upsampling.

1) YUV mean downsampling and converting to RGB format

The average downsampling process is performed on the YUV data to obtain an image in the YUV format after downsampling, and the image in the YUV format is converted into an image in RGB format.

2) Solve the problem of black borders on sudden changes

In order to speed up the processing speed, mean value downsampling and bilinear interpolation upsampling are used, but it will bring certain side effects, especially the edge of the mutation will have excessive defogging, that is, the appearance of black edges.

In view of this situation, the present invention performs bilinear interpolation up-sampling after performing the maximum value filtering process with a radius of 1 on the transmittance in the algorithm, which can effectively avoid the black edge phenomenon.

3) Bilinear interpolation upsampling

Upsampling is divided into proximity upsampling, bilinear interpolation upsampling and bicubic interpolation upsampling. Proximity upsampling is to use the closest pixel value in a block as the value of all points in the block. Bicubic interpolation and upsampling have the best effect, but the complexity of the algorithm is relatively high, so it is not suitable for use in the present invention. In the present invention, bilinear interpolation is used for upsampling, which is to perform linear interpolation between several points, so that the image after upsampling changes smoothly, and the blocking effect can be eliminated. The principle of bilinear interpolation is shown in Figure 1:

In Figure 1, 1, 2, 3, and 4 represent the pixel values of the small image respectively, and the other blank parts are the pixel values to be up-sampled and to be interpolated. Take any point in the first block, and calculate the value of this point. The value of this point is a weighted average of the pixel values of the four points closest to it (corresponding to the pixel values of points 1, 2, 3, and 4 in the figure). The weighting coefficient is related to the distance, the closer the distance is, the larger the weighting coefficient is, and the farther the weighting coefficient is, the smaller it is. Calculated as follows:

dest=a*value1+b*value2+c*value3+d*value4

Among them, dest represents the pixel value to be interpolated, valua1, valua2, valua3, and valua4 represent the values of four adjacent points respectively, and a, b, c, and d represent weighting coefficients, which can be calculated according to the following formula:

$\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; =</span>\frac{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; r</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; t</span>}\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; o</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; *</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; r</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; t</span>}\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; o</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; )</span>}{{\mathrm{<span\; class="notranslate">\; ratio</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}$

$\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; *</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; r</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; t</span>}\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; o</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; )</span>}{{\mathrm{<span\; class="notranslate">\; ratio</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}$

$\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; =</span>\frac{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; r</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; t</span>}\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; o</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; c</span>}}{{\mathrm{<span\; class="notranslate">\; ratio</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}$

$\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; c</span>}}{{\mathrm{<span\; class="notranslate">\; ratio</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}$

In the above formula, ratio represents the magnification of upsampling, and (r, c) in the above formula numerator represents the coordinates of the pixel to be obtained in the ratio*ratio block.

As shown in Figure 2, the embodiment of the present invention is exemplified as follows:

1. YUV average downsampling into RGB small image

The present invention is aimed at high-definition traffic video, such as a video with a size of 500W, and the image size of each frame is 2432*2048. Firstly, mean downsampling is performed on the YUV data. Here, taking n times downsampling as an example, the mean value processing is performed on the Y component and the UV component respectively. Find the mean value in the n*n window without saving it, and directly convert it into an RGB image according to the conversion relationship between YUV and RGB. The conversion formula is as follows:

R=Y+1.371*(V-128)

G=Y-0.698*(V-128)-0.336*(U-128)

B=Y+1.732*(U-128)

Among them, Y, U, and V represent the values of the three components of YUV, respectively, and R, G, and B represent the values of the three channels of RGB after conversion.

2. Obtain the transmittance based on the method of dark channel prior

The present invention is based on the physical model of the foggy image, and the physical model of the foggy image can be expressed as:

I(x)=J(x)t(x)+A(1-t(x))

Among them, I(x) is the existing image (the image to be dehazed), J(x) is the haze-free image to be restored, A is the global atmospheric light component, and t(x) is the transmittance.

2.1. Obtain dark channel

In most non-sky local regions, some pixels will always have at least one color channel with a very low value. In other words, the minimum value of light intensity in this area is a very small number. First find the minimum value in the RGB component of each pixel. The formula for finding the dark channel is as follows:

${\mathrm{<span\; class="notranslate">\; J</span>}}^{\mathrm{<span\; class="notranslate">\; d</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; r</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; \&Omega;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; (</span>\underset{\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; \&Element;</span><span\; class="notranslate">\; \{</span>\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; g</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; \}</span>}{\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; J</span>}}^{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>$

where J ^c represents the pixel value of each channel of the color image, and Ω(x) represents a window centered on pixel x. The theory of the dark channel prior states: J ^dark (x)→0.

2.2. Calculate the atmospheric light value

According to the physical model of the foggy image, in order to restore the fog-free image, the premise is to know the atmospheric light value A. The method for obtaining the A value in the present invention is: get the first 0.01% of the pixels according to the size of the brightness from the dark channel image; in these positions, find the corresponding point from the original foggy image I and calculate the mean value, as the A value .

2.3. Calculate the estimated transmittance

From the deformation of the foggy image, combined with the theory of the dark channel prior, the transmittance can be derived expression, as follows:

$\stackrel{<span\; class="notranslate">\; ~</span>}{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\underset{\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; \&Omega;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; (</span>\underset{\mathrm{<span\; class="notranslate">\; c</span>}}{\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; no</span>}}\frac{{\mathrm{<span\; class="notranslate">\; I</span>}}^{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>}{{\mathrm{<span\; class="notranslate">\; A</span>}}^{\mathrm{<span\; class="notranslate">\; c</span>}}}<span\; class="notranslate">\; )</span>$

in is the transmittance, I ^c represents the value of the three channels, and A ^c represents the atmospheric light value;

2.4. Guided filtering

Guided filtering is performed on the above estimated transmittance, As the input image, the grayscale image is used as the guide image, and the refined transmittance small image is obtained after processing.

3. Perform maximum value filtering for the small transmittance map

The defogging algorithm based on the dark channel prior has such rules: the whiter area needs a higher degree of defogging, and the corresponding transmittance will be relatively small; the darker area (such as the car body or billboard) needs a higher degree of defogging Low, so the corresponding transmittance is relatively high. If there is a darker billboard or car body in a large sky area, then there will be a side effect of blackening the edge of the billboard or car body and the sky area. Assuming that the sky area transmittance t1=0.5 and the billboard area transmittance t2=0.9, then a number between 0.5 and 0.9 will be inserted between t1 and t2. Since the image reconstruction formula is: J=(I-A)/t+A, the pixel value I of the vehicle body or billboard is relatively small, and A is the atmospheric light value (larger, generally above 200), if t is slightly smaller If , it will make the value of J less than zero, so it will turn black.

In order to solve this problem, in the present invention, the maximum value filtering process with a radius of 1 is performed on the transmittance of the small image, which can avoid the appearance of black borders. Of course, this will bring a slight halo effect, but this effect has nothing to do with the problem of black borders Insignificant in comparison.

4. Bilinear interpolation upsampling

For the small transmittance map obtained in step 3, perform bilinear interpolation and upsampling. In this example, 16 times bilinear interpolation upsampling is performed, that is to say, 15 points are inserted between two points of the transmittance of the small image, so that the length and width of the interpolated transmittance image are equal to those of the original image. However, it is necessary to pay attention to the edge processing during interpolation to prevent the transmittance after upsampling from being not in one-to-one correspondence with the original image, and some edge effects will appear in the processing results.

5. Reconstruction of haze-free images

From the transmittance of the large image obtained in step 4, the haze-free image can be reconstructed next. In the present invention, the intermediate conversion process is directly omitted, and the hazy image in the YUV format is directly reconstructed into a YUV haze-free image. calculate first

$\left\{\begin{array}{c}{\mathrm{<span\; class="notranslate">\; A</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0.114</span>}<span\; class="notranslate">\; *</span>{\mathrm{<span\; class="notranslate">\; A</span>}}_{\mathrm{<span\; class="notranslate">\; b</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 0.587</span>}<span\; class="notranslate">\; *</span>{\mathrm{<span\; class="notranslate">\; A</span>}}_{\mathrm{<span\; class="notranslate">\; g</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 0.299</span>}<span\; class="notranslate">\; *</span>{\mathrm{<span\; class="notranslate">\; A</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}\\ {\mathrm{<span\; class="notranslate">\; A</span>}}_{\mathrm{<span\; class="notranslate">\; u</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0.5</span>}<span\; class="notranslate">\; *</span>{\mathrm{<span\; class="notranslate">\; A</span>}}_{\mathrm{<span\; class="notranslate">\; b</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 0.332</span>}<span\; class="notranslate">\; *</span>{\mathrm{<span\; class="notranslate">\; A</span>}}_{\mathrm{<span\; class="notranslate">\; g</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 0.168</span>}<span\; class="notranslate">\; *</span>{\mathrm{<span\; class="notranslate">\; A</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 128</span>}\\ {\mathrm{<span\; class="notranslate">\; A</span>}}_{\mathrm{<span\; class="notranslate">\; v</span>}}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 0.082</span>}<span\; class="notranslate">\; *</span>{\mathrm{<span\; class="notranslate">\; A</span>}}_{\mathrm{<span\; class="notranslate">\; b</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 0.418</span>}<span\; class="notranslate">\; *</span>{\mathrm{<span\; class="notranslate">\; A</span>}}_{\mathrm{<span\; class="notranslate">\; g</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 0.5</span>}<span\; class="notranslate">\; *</span>{\mathrm{<span\; class="notranslate">\; A</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 128</span>}\end{array}\right.$

In the above formula, A _b , A _g , and Ar represent the atmospheric light values of the three channels B, G, and _R , respectively, and A _y , A _u , and A _v represent the atmospheric light values corresponding to Y, U, and V.

Since A _b A _g A _r in one frame of image is the same, the above calculation only needs to be calculated once per frame. Then reconstruct the haze-free YUV image:

$\left\{\begin{array}{c}{\mathrm{<span\; class="notranslate">\; Y</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; Y</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; A</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; A</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}}\\ {\mathrm{<span\; class="notranslate">\; u</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; A</span>}}_{\mathrm{<span\; class="notranslate">\; u</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; A</span>}}_{\mathrm{<span\; class="notranslate">\; u</span>}}\\ {\mathrm{<span\; class="notranslate">\; V</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; V</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; A</span>}}_{\mathrm{<span\; class="notranslate">\; v</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; A</span>}}_{\mathrm{<span\; class="notranslate">\; v</span>}}\end{array}\right.$

In the above formula, Y, U, V represent the data of the original image, t(x) represents the transmittance, A _y , A _u , A _v represent the atmospheric light value corresponding to Y, U, V, Y', U', V' represents image data after reconstruction.

The processing results are analyzed as follows:

The method proposed by the present invention takes about 30ms to process a YUV image with a size of 2432x2048 on a PC, which meets the real-time requirement. And it has a better processing effect, as shown in Figure 3, where Figure 3(a) is the original image, and Figure 3(b) is the image after dehazing before performing the maximum value filter on the transmittance of the small image, so at the edge of the car body The black border problem occurs, as shown in the circle in Fig. 3(b). Figure 3(c) is the haze-free image reconstructed after maximum filtering for the small image transmittance. It is obvious that the problem of black borders will no longer appear after the maximum value filter is added to the transmittance of the small image, and the defogging ability will not be affected.

Claims (1)

1. the rapid defogging method improved, it is characterized in that, the method step is as follows:

Step 1), input the image of a frame yuv format;

Step 2), yuv data average is down-sampled, and is converted into rgb format;

The down-sampled process of average is carried out to yuv data, obtain down-sampled after the image of yuv format, and the image of yuv format is converted to the image of rgb format;

Step 3), ask for little figure transmissivity based on dark primary priori principle;

For step 2) in the little figure of RGB, ask for dark by mini-value filtering, estimation air light value, calculates the transmissivity estimated, finally carrying out Steerable filter to estimating transmissivity, trying to achieve the transmissivity become more meticulous;

Step 4), maximal value filtering;

The liter sampling carrying out bilinear interpolation after radius is the maximal value filtering process of 1 is again carried out, to avoid the phenomenon of black surround to little figure transmissivity;

Step 5), bilinear interpolation rise sampling;

Adopt bilinear interpolation to rise sampling, between several point, carry out linear interpolation, make the image change after liter sampling mild, blocking effect can be eliminated;

Step 6) directly revert to YUV image without mist;

Utilize step 5) the large figure of transmissivity after interpolation, carries out the reconstruct of mist elimination image, directly adopts yuv format to be reconstructed, eliminate the Time and place that middle YUV and RGB changes;

Step 7) input next frame YUV image.