Underwater image restoration is the digital image processing technique which improves the visual quality of images captured under different underwater scenes.
In recent years, this research area has increased attention since improving the visibility, contrast, and colors of underwater images are of significance for many computer applications.
In recent years, this research area has increased attention since improving the visibility, contrast, and colors of underwater images are of significance for many computer applications.
Generally, underwater images are degraded because of light that mainly absorbed and scattered by three water constituent particles:
1. Micro phytoplankton.
2. Colored dissolved organic matter.
3. Non-algal particles.
1. Micro phytoplankton.
2. Colored dissolved organic matter.
3. Non-algal particles.
When the light propagates in an underwater environment, the light received by a camera is mainly composed of three kinds of light which are direct light, forward scattering light, and backscattering light. The received light by a camera faces the problem of color deviation due to the wavelength-dependent light absorption. In general, the red light first disappears with the increasing distance from objects, followed by the orange light, yellow light, purple light, yellow-green light, green light, and blue light. This is the main reason because of which underwater images are dominated by the bluish or greenish tone. Hence, to improve the visual quality of the underwater image, a method that can remove the effects of backscattering light and wavelength-dependent light absorption is needed. To solve this problem, a variety of methods have been proposed, one of them is as follows.
An RGB YCbCr Processing method (RYPro)
This method is proposed for underwater images commonly suffering from low contrast and poor color quality because of light absorption. Moreover, as the depth increases, different colors are absorbed by the surrounding medium depending on the wavelengths in water. Blue/green color is dominant in the underwater ambiance. Color equalization is a widely used approach for underwater image enhancement. In the present work, the processing sequence includes noise removal using linear and non-linear filters followed by adaptive contrast correction of the RGB and YCbCr color planes.
In this method to reduce color cast Beer’s law is used. Beer’s law is commonly employed to correct the pixel intensity by calculating the amount of light absorbed in water. In this method, missing wavelengths are calculated assuming that all the underwater objects are at the same depth.
The purpose of the RGB YCbCr Processing (RYPro) is to improve the visual quality of the image by integrating color cast with subsequent contrast enhancement. In the proposed method, instead of conventional blind color equalization, noise reduction using linear and non-linear filters, color enhancement in the RGB color space, and lastly enhancement in the YCbCr color space are used.
Fig. 1 Flow chart of the sequence of operations in the
proposed RYPro method.
As shown in figure 1, the basic processing sequence of the proposed method are – noise reduction using linear and non-linear filters (which nullify the effect of noise introduced by scattering), adaptive contrast correction in the RGB color space (it neutralize the low contrast and color cast effects due to scattering and refraction, and absorption, respectively) and enhancement of the luminance component (Y) of the YCbCr color space (which diminish the effects of low brightness due to absorption).
1) Noise reduction using linear and non-linear filters: Back scattering of light in the underwater channel effectively introduces noise on the captured image. The effect of such noise is typically mitigated by implementing a suitable linear and/or non-linear filter. In this method a two dimensional (2D) Weierstrass transform having kernel size of 13 × 13 and standard deviation (σ) of 0.5 is adopted for linear noise reduction (noise reduction using linear filters). However, noise reduction with linear filters do blurring of the noisy edge pixels. To alleviate the backscattering noise (associated with region pixels), anisotropic and median filters are used in the proposed algorithm (i.e. noise reduction using non-linear filters). Anisotropic diffusion is given as
δI/δt= div(c(x, y, t)∇I) = ∇c · ∇I + c(x, y, t) I, (1)
where ∇ denotes the gradient, ÷(··) is the divergence operator on the parameters in the parentheses and c(x, y, t) is the coefficient of diffusion. The term c(x, y, t) controls the rate of diffusion and is usually chosen as a function of the image gradient to preserve the edges of the image under test. Motivated by this, as a final step a median filter is used for reduction of speckle and noise is defined in following equation
I (x, y) = median [I (x − 1, y − 1), I (x − 1, y), I (x−1, y+1), I (x, y−1), I (x, y), I (x, y+1), I (x + 1,
y − 1), I (x + 1, y), I (x + 1, y + 1)].
2) Enhancement in RGB color space: Underwater images have low contrast alongside showing a tendency for one color dominancy and low sharpness. These channel impairments can be diluted via methodical processing in the RGB color space after noise removal. Figure 2 shows the sequence of operations in the RGB color space adopted in the proposed method for this purpose.
 |
| Fig.2 Sequence of operations in RYPro method in RGB color space. |
(a) Contrast stretching: Due to low contrast of underwater images, the dynamic range of the histogram is quite less. Contrast stretching is therefore used to redistribute the pixel values between 0 and 255. A contrast stretching algorithm uses a linear scaling function of the pixel
values as IN(x, y) = (I (x, y)−IMin)×((IdMax − IdMin)/(IMax − IMin))+ IdMin.
where IN(x, y) = normalized pixels intensity values after contrast stretching,
I (x, y) = pixel intensity values before contrast stretching,
IMin = lowest intensity of the parent images,
IMax = highest intensity of the parent images,
IdMin = minimum pixel intensity in desired ranges,
IdMax = maximum pixel intensity in desired ranges.
(b) Contrast limited adaptive histogram equalization: Contrast stretching, histogram equalization is considered as a next step for this algorithm. In the proposed approach, the well-known Contrast Limited Adaptive Histogram Equalization (CLAHE) algorithm is used. The CLAHE algorithm can be for images with lighter as well as dark portions. It equalizes the histogram of different sections of an image and exploits the intensity normalized pixels to redistribute lightness values, thus improving local contrast of the image to extract the required hidden information. The procedure adopted for CLAHE operation is as shown in figure3. If Pk denotes the frequency of the occurrence of a pixel’s intensity value that is associated with a bin for each possible intensity, then Pk is defined as Pk = nk; k = 0, 1, 2, . . , 255. where nk is the number of pixels associated with kth intensity value.
 |
| Fig. 3 Sequence of operations in CLAHE. |
As a next step, the cumulative distribution function (cdf) of each of the intensity values is estimated as
where x = 1, 2, 3, . . , M(number of rows), y = 1, 2, 3, . . , N(number of columns). The histogram equalized image I '(x, y) is given as
I'(x, y) = (f loor)((cdfI (x,y) − cdfmin)/(M × N − cdfmin))× 255.
Where cdfmin is the minimum cdf value for the corresponding section. A comparative result obtained by applying conventional histogram equalization method and CLAHE on an underwater image is shown in the figure 4. In the figure 4, the x-axis of the histogram (original and enhanced images) ranges from 0 to 255. Clearly, CLAHE, generates better visual images than that obtained utilizing the conventional histogram equalized method.
Fig. 4 (a) Original image (b) Histogram of original image
(c) Conventional histogram equalized image (d) Histogram of
image in (c) (e) CLAHE operation on original image (f) Histogram
of image in (e).
3) Color preserving processing in YCbCr color space: By processing only the luminance component of an image and leaving the chromatic components undisturbed, image can preserve colors in the enhanced image. Contrast enhancement of the luminance component improves the brightness of an image. For this purpose, the popular YCbCr color model used in digital video and photography has been considered for transform domain operation in the proposed method (as shown in figure 5).
 |
| Fig. 5 Sequence of operations in YCbCr domain for RYPro |
In this model, Y, Cb and Cr represent the luminance, the difference between the blue component and a reference value, and the difference between the red component and a reference value, respectively. By virtue of the property of the YCbCr model, enhancing the luminance component automatically enhances the brightness of the images, leaving Cb and Cr untouched. This improves underwater image quality, which commonly suffer from both low and non-uniform brightness issues.Hence, for this reason that the RGB image is first transformed into the YCbCr domain using for subsequent operation in the transform domain.
After transforming into the YCbCr plane, the contrast stretching method is applied to the luminance component. In order to alleviate the noise introduced in the process, median and Laplacian filters are employed subsequently for image smoothening and edge sharpening, respectively. After the enhancement of the luminance component in the YCbCr plane, the enhanced image is transformed back to the RGB domain, which will be result of the proposed method.
Fig 6 Stepwise images: (a1,a2) Original image; (b1,b2) After noise removal; (c1,c2) After contrast stretch; (d1,d2) After applying CLAHE; (e1,e2) RGB to YCbCr transformed image; (f1,f2) After contrast stretching of luminance component; (g1,g2) After noise removal of luminance component; (h1,h2) YCbCr to RGB transformed – Final image.
Figure 6 shows the stepwise results obtained by applying RYPro on Hammer (a1–h1) and Wheel (a2–h2) images captured from 20.574 cm.
This is how RGB YCbCr Processing method can be used for underwater image restoration.For all above information we have referred the following references:
[1] S. G. Narasimhan and S. K. Nayar, “Chromatic framework for vision in bad weather,” in Proceedings of the IEEE Conference on Computer Vision & Pattern Recognition (CVPR ’00), vol. 1, pp. 598–605, Hilton Head Island, SC, USA, June 2000.
[2] R. T. Tan, “Visibility in bad weather from a single image,” in Proceedings of the 26th IEEE Conference on ComputerVision and Pattern Recognition (CVPR ’08), pp. 1–8, June 2008.
[3] R. Fattal, “Single image dehazing,”ACMTransactions onGraphics, vol. 27, no. 3, article 72, pp. 1–9, 2008.
[4] K. He, J. Sun, and X. Tang, “Single image haze removal using dark channel prior,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 33, no. 12, pp. 2341–2353, 2011.
[5] Sehgal, A., Kadarusman, J., and Fife, L.D. (2004). Touch: A robotic vision system for underwater object tracking. In IEEE Conference on Robotics, Automation And Mechatronics, volume 1, 455–460.
[6] Sural, S., Gang, Q., and Pramanik, S. (2002). Segmentation and histogram generation using the hsv color space for image retrieval. In International Conference on Image Processing, volume 2, 589–592.
Thankyou!!
