Advanced Dehazing Techniques for Image Restoration

Objective

A fast single dehazing method used to improve the image quality and speed of image restoration.

Abstract

The atmospheric scattering and absorption give rise to the natural phenomenon of haze, which severely affects the visibility of scenery. Dehazing is the method used to remove the haze. In recent year, many works have been done to improve the visibility of image taken under bad weather. The images that are captured under hazy conditions suffer from color distortion and attenuation. By estimating the transmission map and atmospheric light the a fast dehazing algorithm is proposed based on the Dark Channel Prior theory and gray projection.

The algorithm reduces operational complexity and optimized average filtering method is used.

An atmospheric light acquisition method is designed based on the regional projection and an amendment strategy for large white areas is proposed. In the restored images to solve the problem of low brightness an adaptive adjustment method based on human visual perception is proposed. Relative to current state-of- the-art algorithms, it is concluded that the images restored from the proposed method appear clearer and more natural.

Get quality help now

Prof. Finch

Verified writer

Proficient in: Technology

4.7 (346)

“ This writer never make an mistake for me always deliver long before due date. Am telling you man this writer is absolutely the best. ”

+84 relevant experts are online

Hire writer

The algorithm has a wide range of applicability to ensure the balance of quality and the speed of the image restoration. This algorithm can restore images to a clear and natural state and ensure the balance of quality and the speed of image restoration. It is also highly efficient method since it can process large images.

Introduction

Humans are able to see the objects as light is reflected from them.

The reflected light is absorbed or scattered by the air. If the air contains additional particles like fog, mist, the visibility of the object reduces. Thus original light is replaced by the reflected light. This results in hazed image. Dehazing is the process of removing the haze. Various methods are their to improve the image quality of the hazed image. A Fast single dehazing method is proposed based on dark channel prior theory and gray projection. In the existing methodologies, the quality of restored image is poor and needs a large computation time. According to atmospheric scattering model, the hazy image consists of both the attenuation part and scattering part. The hazy image can be expressed as

I (x) = J (x)t(x) + A (1 − t(x))

Where I (x) is the input hazy image, J (x) is the haze-free image, A is the atmospheric light and t(x) is the transmission rate of the medium. The purpose of image dehazing is to recover J (x) from I(x).

According to Dark channel prior, atleast one color channel of an RGB image has some pixels of low intensities, which tend to zero. This method cannot be applied to white areas, such as sky areas and under water surfaces. The proposed method is based on gray projection and physical model. The speed of computation is fast and the quality of the restored image will be high. This method is divided into three steps. Firstly, estimating the transmission map using minimum filtering and fast average filtering. Secondly, to estimate the atmospheric light the gray projection technique is used. Thirdly, the image is restored based on the Webner Fechner Law, to adjust the brightness of the image.

For calculating the atmospheric light, the pixels with large brightness is taken ,i.e, sky areas and the gray projection method is used for finding the location with high brightness level. The horizontal projection of the filtered image is calculated and sum of horizontal projections within a width of 2b+1 is taken and select the maximum region. Next, The vertical projection of the filtered image is calculated and sum of vertical projections within a width of 2b+1 is taken and select the maximum region. The maximum region is saved as R(x,y). Finally, the average gray value of top 1% pixels are arranged in descending order and the pixels with the largest brightness is selected. The value of A is expressed as follows:

A = mean (max0.1 R(x))

The region is projected. From projected image, the repeated calculation is performed over a certain region by using box filter acceleration algorithm to increase the speed of computation.

Calculation of Atmospheric Light

For calculating the atmospheric light, the pixels with large brightness is taken i.e., sky areas and the gray projection method is used for finding the location with high brightness level. The horizontal projection of the filtered image is calculated and sum of horizontal projections within a width of 2b+1 is taken and select the maximum region. Next, The vertical projection of the filtered image is calculated and sum of vertical projections within a width of 2b+1 is taken and select the maximum region. The maximum region is saved as R(x, y). Finally, the average gray value of top 1% pixels is arranged in descending order and the pixels with the largest brightness is selected. The value of A is expressed as follows:

A = mean (max0.1 R(x))

The region is projected. From projected image, the repeated calculation is performed over a certain region by using box filter acceleration algorithm to increase the speed of computation.

Methodology

The methodology centers on the atmospheric scattering model, which represents a hazy image I(x) as a combination of the direct attenuation and air light scattering components. The model is expressed as:

I(x)=J(x)t(x)+A(1−t(x))

where I(x) is the observed intensity, J(x) the original intensity, A the atmospheric light, and t(x) the transmission medium.

To overcome the challenges posed by white areas and ensure detailed restoration, the study introduces an innovative atmospheric light acquisition method using gray projection and an adaptive brightness adjustment strategy based on human visual perception.

Implementation Aspects

For the implementation of the proposed system, both the principles of dark channel prior theory and the improvement statergies are used. The steps of implementatin are as follows:

Step 1: The hazy image I (x) is input.

Step 2: Minimum Filter is used on I (x) to obtain the DCP image M(x).

M(x) = min c ɛ{r,g,b}( Ic(x))

Step 3: M(x) is operated on by an averaging filter to obtain the smoothed image

Mavg(x).

Mavg(x) = averageλ(M(x))

Step 4: Mavg(x) is compensated by a grayscale filter to obtain the corrected image

D(x).

D(x) = min(A*Mavg(x),M(x))

Step 5: The atmospheric lighting information is automatically obtained using the projection method, and the atmospheric light intensity A is generated.

Step 6: The initial transmittance t^(x) is calculated.

t֮ = 1-ὠD(x)/A

Where the correction factor is always greater than zero.

Step 7: The transmission map is adaptively modified using the segmentation mechanism to obtain t(x).

Step 8: The image is restored based on the physical model , to obtain the restored image J(x).

J(x) = I(x) -A/max(t(x),to) +A

Step 9: The brightness compensation is performed on the restored image and Jd is obtained.

Jd = J(255+k)/J+k

Where k is the adjustment coefficient. If the value of adjust coefficient is small then the degree of adjustment will be high.

Step 10: The image Jd is the output.

Comparison of Proposed Model With Other Methods

For the comparison, the relative time taken for computation is calculated. For the same image the computation time is calculated at different resolution. As the size increase, the processing time also increases. In the objective evaluation, the proposed method is compared with the existing methods. This method evaluates the contrast enhancement of each image before and after dehazing. The quality of the image is described using the three parameters.

e = nr-no/no

r֮ = exp (1/nr ∑𝑃𝑖ɛ𝛹 log 𝑟𝑖)

ϭ = ns/dimx*dimy

Where e is the new visible edge ratio and r is the visible edge normalized gradient and ϭ is the percentage of saturated black or white pixels.

For each different images of different size, the image is tested several times and average value is taken at the output. The relative time approach is used to compare the increasing image size with respect to increasing computing time. It is expressed as

Tr = tn /t1

Where, t1- Processing time of smallest image and n is the image index which increases with the size of images.

Expected Results

The proposed method is used for processing large images and it can compute fastly than other methods. The He2 method is faster but it fails to recover the true scene radiance of the image and it remains bluish.

Conclusion

The fast single-image dehazing method based on the physical model and gray projection presents a significant advancement in image processing, offering a balance between image quality and computational speed. This method is particularly effective for large images, ensuring both clarity and natural color restoration with high efficiency.

References

1. Wencheng Wang, Tao Ji, Faliang Chang, "A Fast Single-Dehazing Method Based on a Physical Model and Gray Projection," IEEE Access. 2794340, Feb. 2018.
2. K. He, J. Sun, and X. Tang, "Single image haze removal using dark channel prior," IEEE Trans. Pattern Anal. Mach. Intell., vol. 33, no. 12, Dec. 2011.
3. Q. Zhu, J. Mai, and L. Shao, "A fast single image haze removal algorithm using color attenuation prior," IEEE Trans. Image Process., vol. 24, no. 11, Nov. 2015.