"Fog technology" makes monitoring no longer "seeing flowers in the fog"

Removing fog from video and improving image quality is a key technology for enhancing the value of outdoor video surveillance systems. This paper analyzes the defogging technology by establishing the relationship between the image blurring technique and the physical model of the fog image, synthesizes the basic principles of the image dehazing method proposed in recent years, and explains how some typical dehazing methods visually appear. The effect of defogging.

First, the haze has caused the development of desmear technology

Recently, PM2.5, a professional vocabulary in the field of meteorology, has become a topic of public concern. Droplets and small solid particles in the air not only endanger human health, but at the same time, due to scattering of a large number of suspended particles, the visibility of the atmosphere decreases, and the color and contrast of outdoor images deteriorate. Affecting the extraction of information in images, it also leads to the decrease of outdoor sharpness, which leads to frequent traffic accidents. Therefore, image dehazing technology has become an important topic in the field of image processing and computer vision research, and it is also one of the issues that people urgently solve.

In recent years, with the continuous development of technology, it has become possible to perform defogging processing on scene images in foggy weather, and there have also been significant advancements in the clarity and realism of defogging images. Image dehazing technology is an important content in the field of image processing and computer vision. Its main application areas are video surveillance, terrain survey, automatic driving and target tracking. In the application process, the authenticity of the image after defogging and the real-time processing have become the focus of research and attention on dehazing technology.

Second, the image fog technology

There are many ways to process images using digital image processing techniques. In a nutshell, there are still two traditional types: one type of image enhancement and the other type of image restoration. The image enhancement method is based on the characteristics of low brightness and low contrast of image rendering, highlighting certain information in the image according to specific needs, and weakening or removing some unnecessary information to complete; the image restoration is from the weather based on the physical model. The degraded image restoration method performs modeling and analysis of atmospheric scattering from the perspective of physical origin to realize scene restoration. In the recovery process, various parameters in the model are generally used to estimate the parameters in the model first, and finally the equations are solved to calculate a clear image. Compared with the two methods, based on the reconstructed defogging algorithm, the dehazing is achieved in principle, the fog is estimated more accurately, and the clear image before the fog can be truly restored. The targeted defogging effect is natural, and generally there will be no Loss of information.

1, image processing based image enhancement technology more application

Typical fog image enhancement methods include grayscale histogram transform method, frequency domain analysis method, and Retinex algorithm based on color consistency.

The grayscale histogram transform method transforms the histogram of the foggy image into an evenly distributed form, thereby increasing the dynamic range of the pixel gray value to achieve the effect of enhancing the overall contrast of the smog image.

The wavelet analysis algorithm in the frequency domain is a multi-scale image enhancement algorithm based on wavelet analysis. Its main idea is that the energy of the fog is mainly concentrated in the low frequency part of the image and has a small impact on the high frequency part. Therefore, based on this, the image's The high and low frequencies are enhanced or attenuated, respectively, to achieve clear images. It is worth mentioning that there is currently a new multi-scale analysis method based on wavelet transform - curve wave analysis algorithm, which is particularly suitable for signal processing of anisotropic singularity features, and well compensates for wavelet transforms in images. The limitations of the curved edge enhancement can take advantage of the curvelet transform and use the vanishing point detection based on the curved wave to automatically defogging fog images.

The color consistency or color constancy theory is based on human visual characteristics and proposes a theory that human visual system can ignore the change of illumination in the environment and obtain a stable color perception. Retinex algorithm is a kind of image enhancement method based on color constancy theory. The algorithm enhances the weakened illumination due to fog interference through the feature of color invariance of vision system, so as to achieve the purpose of image enhancement.

2. The image restoration method based on the weather physics model becomes a new trend

(1) Method for synthesizing scene depth models for multiple images

The early restoration algorithm uses the scene depth to solve the atmospheric scattering equation to obtain a clear image. Then a method for synthesizing the depth model of the scene using multiple images of the same scene under different weather conditions has been achieved, and good results have been obtained, but subjective Due to the limitation of conditions, the estimated depth of the scene is often not accurate enough. At the same time, due to lack of sufficient prior conditions to be constrained, it is easy to cause the result of recovery to be inconsistent with reality.

(2) Analysis of polarization angle of light and partial differential equation

Later, people analyzed the fog from the perspective of the polarization of light, separated the light scattered by the fog into horizontally polarized light and vertically polarized light using multiple images of the same scene, and designed a corresponding filter to eliminate fog and light. Impact, to achieve the purpose of defogging, this method has obvious defogging effect, image distortion is small, but the amount of calculation is too large, it is difficult to apply to the actual. In addition, partial differential equations are also widely used in the field of image restoration in fog days. The main method is to establish the gradient field corresponding to the atomization image based on the atmospheric scattering model, and then construct partial differential equations based on the relationship between image depth and gradient. The solution to obtain a clear image, the method can achieve a single image of the blind defogging, but the process of constructing and solving partial differential equations is cumbersome and equally difficult to achieve.

(3) Single Degraded Image Fog Concentration Analysis

In recent years, many researchers have been working on how to achieve a complete defogging effect for a single degraded image according to the change in fog concentration in the graph: Statistics show that fog-free images must have high contrast with respect to fog images, so we The purpose of defogging can be achieved by maximizing the local contrast of the restored image. This method can greatly enhance the contrast of the image, but it easily leads to color distortion of the image, and an aperture effect may occur at a place where the depth of the scene is not continuous. Figure 1.

Figure 1 Tan method of defogging effect

(4) Reflection estimation method of light

Because the reflectivity of the object surface is fixed and has nothing to do with the light intensity of the surface, it can also use the object's reflection of the light to estimate the transmittance of the light to achieve the purpose of defogging, but the algorithm requires the image to have different colors locally. Therefore, when the concentration of fog is large and the image is close to white, the corresponding parameters cannot be estimated, resulting in failure of defogging. Moreover, this method is only effective for color images, and the amount of calculation is large. The experimental results are shown in Figure 2.

(5) Single image dehazing algorithm based on dark channel prior

The single image dehazing algorithm based on the dark channel prior is a simple and effective image restoration method proposed in 2009. This method is based on a dark channel priori that is common in nature, and is used for sunny days. There are always some "dark spots" in the outdoor image. These "dark spots" have low values ​​of at least one color channel, so when the image is disturbed by fog, these originally low values ​​are affected by the scattered light of the atmosphere. With a sharp increase, these points can be used to estimate the concentration of fog in the shooting scene and restore a clear, fog-free image. However, this method can only roughly estimate the distribution of atmospheric light in the image, and it is necessary to optimize the transmittance by combining the soft ridge map or the bilateral filtering algorithm. The defogging effect of this method is shown in Figure 3.

Third, the dark original color fog method improvement

In the process of perfecting the transmittance, the method of soft mapping is used to repair the underestimated dark pixel values ​​by using the maximum filter at the boundary between the near and far scenes of the dark original image. However, in the calculation process, the soft map has large computational overhead and high time complexity. In the process of solving the linear system, the speed is slow, and the computational efficiency becomes the biggest obstacle to the practical application of this method.

If you select the guide filter that maintains the edge of the image, approximate this erosion process by image-directed filtering, which not only achieves similar results but also reduces the runtime. Compared with the classical bilateral filtering, guided filtering is an explicit filtering, which not only has linear time complexity, but also has better preservation of the image edges, and can also achieve image edge smoothness, detail enhancement, and image fusion denoising, etc. Features.

1. Improved image dehazing experiment results

In order to verify the effectiveness of the improved algorithm, a defogging experiment was performed on the same fog image using image filtering. By observing Figure 4, we can see that in the visual effect, although there is no obvious difference in the optimization results of the initial transmission map using the soft-sweeping map and the guided image filtering, the amount of calculation using the soft-squint diagram accounts for approximately this entire calculation. 90% of the quantity, while the guide image filtering algorithm has low computational complexity, the operation efficiency has been greatly improved, and at the same time, a large amount of memory space is saved. The statistical results are shown in Table 1.

Fourth, the application of fog removal technology in the field of security monitoring

The author's company cooperated with Nanjing University of Posts and Telecommunications to develop smog and video processing technology to deal with the adverse effects of video images due to haze and other weather conditions, and to restore a clearer image for human eye recognition. This processing method was adopted. A unique image restoration algorithm based on the prior art dark channel theory, using the atomization model and prior dark channel can directly estimate the thickness of the fog, and restore a high-quality image of the fog.

With the development of industry and its impact on the climate, haze has increasingly become a common weather phenomenon, which has a great impact on the picture quality of outdoor application monitoring systems. The dehazing technology can improve the quality of video surveillance from multiple perspectives and can be used for fog treatment in various foggy weather conditions; it can significantly improve the contrast of the image, make the image transparent and clear, and can significantly enhance the details of the image. Information, so that the original details of the hidden images are fully displayed; can improve the saturation of the image, so that the image is vivid, lively and vivid, the image after the fog treatment to maintain accurate hue, natural appearance, and thus obtained a good image quality With visual experience.

Therefore, from the perspective of application scenarios, the dehazing technology can be used in a variety of outdoor applications, such as in the field of security monitoring, and can greatly improve the performance of existing monitoring systems in bad weather such as foggy days; it can be applied to the field of highway traffic monitoring , It can avoid the outdoor surveillance camera can not see the monitoring object, resulting in the consequences of failure at a critical moment; can also be applied to remote sensing image processing and military technology and other fields.

Figure 5 Application of dehazing technology in traffic

V. Conclusion

The image enhancement method based on image processing has the characteristics of significant contrast enhancement, prominent image detail, and obvious visual effects. This method has been widely used in practice. The image restoration method based on the physical model is highly targeted and the recovery results obtained are natural. It is believed that the technology will surely achieve greater development, especially the method based on the recovery of fog images in dark colors prior. Using bootstrap filtering instead of soft mapping, it breaks through the biggest bottleneck of this method—that is, the problem of large computational overhead and high time complexity.

The future research direction of image dehazing technology will focus on improving its real-time performance and hardware implementation. At the same time, the search for a more complete physical model to describe the complex atmospheric conditions and explore the research on the dehazing algorithm based on these models will be a challenging topic in the near future.