Image defogging method based on rayleigh scattering adaptive color compensation

Abstract

The application discloses a Rayleigh scattering self-adaptive color compensation image defogging method, applied to the technical field of image processing, and comprises the following steps: based on the improved bright channel, reference images are obtained to segment bright and dark areas of a foggy image, a threshold value is set to select the image of a dark channel prior theory failure part, the transmittance is optimized through a tolerance value, the foggy image is smoothed by using a guide filter, and the atmospheric light value is analyzed through the Rayleigh scattering theory; the image is evaluated by using an HCC objective parameter, the atmospheric light value of the B channel is repeatedly iterated and compensated, and finally, the image defogging is completed. The application effectively improves the image defogging precision, enriches the image details, solves the image color difference problem, improves the image brightness, and eliminates the sky area halo problem.

Description

Technical Field

[0001] This invention belongs to the field of image processing and image quality enhancement. This invention relates to techniques such as Rayleigh scattering compensation and segmentation using prior images of bright and dark channels. Background Technology

[0002] With the increasing number of private cars, exhaust emissions have become one of the direct causes of smog. In smoggy conditions, a large number of suspended particles exist in the air. Incident light interacts with these particles multiple times, severely affecting the imaging capabilities of the human eye and cameras, making it impossible for people to accurately obtain target information, and leading to a series of irreversible consequences such as car accidents. Therefore, image dehazing has come into focus and become a popular research area recently.

[0003] To improve image quality and avoid the negative impact of extreme weather on images, scholars have proposed using... The Retinex theory was proposed based on the retinal color theory. The retinal color theory posits that the perceived color and brightness of objects are not solely determined by the incident light entering the eye, but are also inextricably linked to ambient light. Therefore, Retinex is a term derived from the English words for the retina and cortex, aptly expressing its theoretical meaning. The Retinex theory categorizes degraded images into high-frequency and low-frequency images, optimizing the degraded image and improving image contrast by removing the low-frequency image.

[0004] Retinex theory does not consider the root causes of image degradation; while the processed image quality is improved, it is still not ideal. Researchers, through observation of numerous clear outdoor images, discovered that in most clear images, one of the RGB color channels of a pixel always has a low pixel value, tending towards zero. Using a minimum value function window to filter the image, the selected pixels all tend towards zero; these are then stitched together to form an image called the dark channel image. Because fog acts like a white veil covering clear images, increasing the pixel values ​​and making the image appear whiter and less detailed, the presence of fog in foggy images can be clearly observed in the dark channel image. Atmospheric light values ​​and transmittance can be calculated from the dark channel image, allowing for the restoration and repair of foggy images. To further improve the optimized images, researchers observed a large number of images in the PASCAL VOC 2012 dataset and found that in most outdoor images, there is always one color channel with a higher pixel value among the three RGB color channels, which tends to be 255. This idea is similar to the dark channel prior theory. By filtering through a maximum value function window, the pixels selected with a pixel value tending to be 255 are stitched together to form an image, which is called the bright channel image. This theory should mostly be used in conjunction with the dark channel prior theory.

[0005] Despite some achievements by scholars, the field of image dehazing still faces the following two challenges: 1. Images processed using traditional dark channel prior theory tend to be darker overall. 2. Processed images perform poorly in the sky and bright areas. These areas introduce a large amount of noise, resulting in noticeable halo effects. 3. Image details are obscured. Details are unclear, or even lost altogether. 4. The image has excessively high color saturation. The overall color of the image is too dark, resulting in color differences compared to the unprocessed image. Summary of the Invention

[0006] I. The image dehazing system model based on Rayleigh dispersion adaptive color compensation for hazy images is as follows: This invention runs on a computer with an Intel(R) Core(TM) i7-6700 CPU @ 3.40GHz and 8GB of system memory. The flowchart is as follows. Figure 1 As shown.

[0007] The defogging process modeling of this invention describes, in part, the incident light entering the imaging tool after interacting with particles in the air; in part, it describes the imaging light entering the imaging tool after being enhanced by other ambient light, which can be described by the following formula: (1) In the formula, For images with fog, The processed image, i.e., the clear image. The transmittance of the image. The atmospheric light value of the image. Represents the position of a pixel in the image.

[0008] The dark channel prior theory selects the top 0.1% of pixels as atmospheric light values. Based on this, this invention uses guided filtering to smooth the image and adaptively adjusts the histogram correlation coefficient (HCC). After determining the atmospheric light values ​​of the three channels, HCC is evaluated by continuously increasing the atmospheric light value of the B channel. Because the atmosphere consumes a large amount of atmospheric light from the B color channel, the HCC initially increases accordingly. However, when it reaches a certain threshold, color distortion occurs, causing the HCC to decrease. Therefore, the threshold is adaptively selected.

[0009] (2) in, Represents atmospheric light value, , , These are the atmospheric light values ​​for the three RGB channels of the image. , , The atmospheric light value before optimization. The function is a function to optimize and compensate for the atmospheric light value of channel B.

[0010] This invention improves the bright and dark images by removing the selection window when selecting the dark channel image and the bright channel image, resulting in an improved channel image. The difference between the improved bright channel image and the dark channel image is then calculated, and the resulting image is named the reference image. Regions with relatively small pixel values ​​are selected, and the transmittance is optimized using Jiang Jianguo's optimization method. After obtaining the optimized transmittance, formula (3) is used for dehazing.

[0011] (3) (4) in, For a clear image, For images with fog, Optimize the tolerance value for Jiang Jianguo. Transmittance, For parameters that are not zero, Atmospheric light value, For reference image, The position of a pixel in the image. It is a function with maximum value. To select a threshold.

[0012] III. Advantages of this invention compared to existing technologies: 1. Compared with existing technologies, this invention corrects the image color difference problem.

[0013] 2. This invention optimizes image details, preserving the original image details and resulting in a better image effect after dehazing.

[0014] 3. This invention solves the problem of poor performance of existing technologies in bright areas of the sky.

[0015] 4. This invention improves the brightness of the image. Attached Figure Description

[0016] Figure 1 This is a flowchart of the present invention. Figure 2 Foggy images Figure 3 Image after DCP processing Figure 4 Image after MAMF processing Figure 5Image after PDE processing Figure 6 TDD image with fog Figure 7 This is a flowchart of the present invention. Detailed Implementation

[0017] To verify the effectiveness of the Rayleigh dispersion adaptive color compensation image dehazing system proposed in this invention, simulated tests were conducted on foggy images, and three dehazing methods were compared. These three methods are: traditional Dark Channel Prior Dehazing Theory (DCP), Model-Assisted Multi-Band Fusion for Single Image Enhancement (MAMF), a Sandstorm Image Enhancement Method Based on Dehaling Dark Channel Prior Dehazing (TDD), and Edge-Driven DCP Image Dehazing Method (PDE). Figures 2-7 As can be seen, the DCP-enhanced image exhibits a noticeable halo effect in the sky, with overall low brightness and excessive depth of field; the MAMF-enhanced image introduces noise into the nearby mountain peaks, causing them to appear washed out, and the defogging at the boundary between distant and nearby peaks is incomplete, resulting in abrupt edges; the TDD-enhanced image also over-magnifies the sunset over the distant snow-capped mountains; the PDE-enhanced image is generally dark, with excessive color saturation in the nearby mountain peaks, affecting the visual effect. The enhanced image of this invention is as follows... Figure 7 As shown in the image, the colors are natural, and the boundaries between distant and nearby peaks are clear and natural. The image quality is even better after dehazing.

[0018] To further demonstrate the effectiveness of this invention, it was evaluated using four parameters: Peak Signal-to-Noise Ratio (PSNR), Structural Similarity (SSIM), Average Pixel Value (AVE), and Universal Quality Index (UQI). PSNR represents the degree of image distortion; a higher PSNR value indicates less image distortion. SSIM calculates similarity based on brightness, contrast, and image structure; a higher SSIM value indicates higher image similarity. AVE represents the average pixel value of the image; a higher AVE value indicates better image brightness. UQI is an objective parameter designed based on the overall image correlation loss, brightness distortion, and contrast distortion; a higher UQI value indicates higher overall image quality. Table 1 shows that the clear images processed by this invention have slightly higher objective parameter values ​​than those processed by existing technologies. To facilitate further observation, the PSNR values ​​of each technology were normalized and the data were averaged. The average value of the objective parameters of the present invention increased by approximately 19.74% compared to DCP, approximately 23.11% compared to MAMF, approximately 19.20% compared to TDD, and approximately 11.64% compared to PDE.

[0019] Table 1 Figure 1 Objective evaluation parameter table Objective parameters DCP MAMF TDD PDE This invention PSNR 62.8733 67.0110 63.9949 67.2597 68.1270 SSIM 0.9949 0.6186 0.7403 0.9976 0.9984 AVE 0.3070 0.3785 0.3897 0.3313 0.4459 UQI 0.6446 0.7780 0.8389 0.7607 0.9582 Mean of each algorithm 0.6438 0.6113 0.6522 0.6905 0.7709

Claims

1. Based on the improved bright channel, a reference image is obtained to segment the bright and dark regions of the foggy image. A threshold is set to select the image of the part of the dark channel that fails the prior theory. The transmittance is optimized by the tolerance value, and the foggy image is smoothed by guided filtering. The atmospheric light value is analyzed by Rayleigh dispersion theory.

2. The image is evaluated using objective HCC parameters, and the atmospheric light value of the B channel is iteratively compensated to finally complete the image dehazing.

Images