Underwater images degraded due to low contrast and visibility issues. Therefore, it is important to enhance the images and videos taken in the underwater environment before processing. Enhancement is a way to improve or increase image quality and to improve the contrast of degraded images. The original image or video which is captured through image processing devices needs to improve as there are various issues such as less light available, low resolution, and blurriness in underwater images caused by the normal camera. Various researchers have proposed different solutions to overcome these problems. Dark channel prior (DCP) is one of the most used techniques which produced a better Peak Signal to Noise Ratio (PSNR) value. However, DCP has some issues such as it tends to darken images, reduce contrast, and produce halo effects. The proposed method solves these issues with the help of contrast-limited adaptive histogram equalization (CLAHE) and the Adaptive Color Correction Method. The proposed method was assessed using Japan Agency for Marine-Earth Science and Technology (JAMSTEC), and some images were collected from the internet. The measure of entropy
Underwater images are very important for discovering new species such as fishes, exploring aquatic life, coral reefs, and aquatic plants, or saving them from extinction. In the beginning, underwater images were mainly used for military and civil applications [
Absorption and light scattering is the main reason for an underwater image to get blurred such as overexposure, uneven illumination, fog, or lack of light [
This paper aims to propose an image processing technique to enhance the image more accurately, remove blurriness, reset color degradation, and contrast adjustments. This paper proposes a framework to restore natural color, adjust dark and bright regions, and most importantly noise removal so information can be extracted from it. This literature proposed a method that would be feasible for real-time application and the proposed technique seems good enough to handle multiple images and return results accurately as compared to other methods discussed in the literature review section. DCP has some issues such as darkening images, reduced contrast, and producing halo brightness regions. These problems were solved by first applying contrast stretching and then passing this image to DCP for enhancement. Finally, if any darkened or halo regions were produced by it then CLAHE was applied to it. CLAHE works best for nearly all types of images; however, manually adjusting the clipping value causes more computational time and bad results.
The paper is organized as Section 2 is for Literature Review; detail of the proposed model is available in Section 3, Section 4 is revered for detailed results and experimental setup, and finally, the paper is concluded in Section 5.
A lot of work has been done in the field of underwater images [
Unnatural coloring and dominance of blue-green color are the two different issues in underwater image enhancement [
To solve this problem a novel approach was presented by [
The method of Li et al. [
Emberton et al. [
Underwater images are widely used for discovering various plants, fishes, and shipwrecks. However, current systems have some limitations such as color diminishing, blur images, low contrast, and unbalanced light conditions. To overcome these issues white balancing technique is used. CLAHE and DCP are used to increase the quality of underwater images and videos. This approach has four steps, converting the input source into grayscale and then applying DCP and CLAHE to improve image contrast than the white balancing technique for further processing.
Initially, input (image or video clip) is passed to the system for processing. Input converted from RGB to grayscale. That grayscale image is transferred to Dark Channel Prior which increases contrast, but the noise remains the same. The image is then passed to a Homomorphic filter which increases the contrast of an image as well as removes multiplicative noise and removes non-uniform light. DCP tends to darken the images which can be resolved by using CLAHE Contrast limited adaptive histogram equalization. However, the results of CLAHE can be improved by adjusting clip value, and manually adjusting these values take up a lot of time. If the clap value is higher than CLAHE will take more computational time and cannot be used for real-time applications. So, it is important to adjust the clip value automatically for real-time applications.
After enhancement, the color cast is checked and if any unnatural color is present, it can be removed with two main techniques-the Adaptive color balancing technique and the fusion technique that is used to restore natural color and remove any bright regions in an image. The adaptive color balancing technique is a way toward expelling unreasonable shading throws of an image, with the aim that the white region of the image should be rendered as white in underwater images. Legitimate camera white equalization needs to consider the “shading temperature” [
The main reason for using that system was that it can be implemented in a real-time application which would be helpful in different fields. A detailed chart of the proposed framework is provided in
This stage typically aims to convert an RGB image or video into a grayscale image or video. Rgb2gray changes over RGB qualities to grayscale values by distributing the weighted whole of the R, G, and B segments [
Above
In above
The main purpose of using a Homomorphic filter and CLAHI rather than one was to correctly enhance the image quality, and if one of them could not enhance it then the other one would be able to do it, so this way it can be applied in nearly all types of images. The homomorphic filter can easily handle noise and non-uniform light conditions which makes it a better technique for underwater images.
Above
Various techniques have been proposed to overcome underwater issues. Underwater images have various issues which include less light available as water is 800 times denser than air which causes a color absorption effect. These things create blurriness, low contrast, color diminishing, and less visibility. Due to these problems in underwater images, research finds it one of the most difficult areas in research. To overcome these issues a new method is proposed in this study. Initially, the white balancing technique was developed to check the effects on underwater images. Fusion produces better results and removes color cast from images. Initial results of input images are given below:
The Contrast Limited Adaptive Histogram Equalization (CLAHE) works on small blocks in the picture that are called tiles, as opposed to the full picture [
For the histogram, H (i) its cumulative distribution H’ (i) is:
To use the above equation as a remapping function, we need to normalize the H’ (i) in such a way that the maximum value should be equal to 255 (or equal to the maximum values of intensity in images).
White balancing is the technique to remove unreasonable shading throws, with the goal that objects which seem white should render white in your photograph. We need to convert the input image to grayscale so we can get the mean luminance and then extract individual red channel, green channel, and blue color channels. After that, we need to calculate the same mean for every channel and then combine the red, blue, and green channels into a single true RGB channel.
Haze-free image is based on the dark channel prior where one or more than one channel of color has some pixels whose intensity levels are low or near to zero in non-sky patches. In the literature, it is observed that several research studies highlight the haze issues in underwater images. So, to overcome this issue this study also applies DCP. A grayscale image was processed and sent to DCP for processing, and it adjusted the intensity of each pixel. For an arbitrary image J, its dark channel is given by:
Human observation is exceptionally touchy to edges and fine subtleties of a picture, and since they are made principally by high recurrence segments; the visual nature of a picture can be gigantically debased if the high frequencies are constricted or finished evacuated. Conversely, upgrading the high-recurrence segments of a picture prompts an improvement in the visual quality [
Fusion is a way of combining two or more images into a single image. This resulting single picture has more details and is more precise than any other single source picture, and it contains all the important information about the image [
The Japan Agency for Marine-Earth Science and Technology (JAMSTEC) collected various pictures and videos from the deep sea for research purposes. To verify the performance of the proposed framework various experiments are performed on the JAMSTEC [
Steps | Performance criteria | |||
---|---|---|---|---|
MOE | EME | MSE | PSNR | |
Gray image | 7.36 | 5.31 | 9.53 | 180.37 |
Dark channel prior | 7.37 | 5.31 | 82.38 | 20.67 |
CLAHE | 7.69 | 16 | 5.31 | 39.85 |
White balancing | 7.62 | 16.44 | 5.311 | 39.88 |
Sharpened images | 7.69 | 20.22 | 5.311 | 39.82 |
Final output result | 7.69 | 5.31 | 44.54 | 26.85 |
Above
To check the performance of the proposed framework following performance parameters are used.
Measure of Entropy (MOE) Measure of Enhancement (EME) Mean Square Error (MSE) Peak Signal to Noise Ratio (PSNR)
Entropy is a statistical parameter that indicates the randomness in image texture. Entropy H (I) of an image is defined as
Here P (i, j) is the possibility of occurring of any pixel’s intensity values with spatial coordinates (i, j). If the randomness in texture is high then variation in pixel intensity is also high. However, the value of randomness is not a characteristic of underwater images and it arises due to the unnatural color present in the underwater environment. This decreases results for entropy and produces lesser entropy for images. Intuitively, once underwater images were enhanced, it would result in higher average information by increasing the intensity variation. To verify the information about an underwater image, the entropy of the enhanced image can be increased for all types of images. The main reason for the underperformance and bad results of the GW and APE methods is that they are mainly based and focused on blind color equalization.
Another quality metric proposed by Weber was a contrast-based measure of enhancement also known as the “no-reference image quality assessment” (NR-IQA) metric to quantify the value of the contrast enhancement obtained in images. EME value also signifies the contrast quality improvement after enhancement; we used it in the present context as a quantitative NR-IQA parameter.
Here CR represents the minimum and maximum intensities in an image block,
The mean squared error (MSE) of an estimator measures the average of the squares of the errors that is, the average squared difference between the estimated values and what is estimated. MSE is also calculated by comparing the original image and the noisy (compressed image) which produced results for MSE. MSE is also known as the risk function that corresponds to the expected value of squared error loss. MSE always remains positive and not zero is mainly due to randomness. Minimum the value of MSE shows minimum error.
In the above equation, M represents the number of rows and N represents the columns in the images.
Peak Signal-to-Noise Ratio, also known as PSNR, is an engineering terminology that represents the ratio of the highest power of a signal and the maximum power of corrupting noise that affects the value and its representation. Many signal values have a wide dynamic range so PSNR can be expressed in terms of the logarithmic decibel scale. Values of PSNR block compute the peak signal-to-noise ratio in decibels among any two input pictures. This ratio is mostly used as a quality measurement between the original and a compressed image. It has been observed that the higher the value of PSNR, the higher the quality of the enhanced image, or reconstructed image. The Mean Square Error (MSE) and the Peak signal-to-noise Ratio (PSNR) are the two mainly used error metrics used for comparison between compression qualities of images. The MSE value represents the total squared error among compressed and original images, whereas PSNR represents a measure of the highest error. A minimum value of MSE shows the minimum error, and the higher value of PSNR shows better image quality.
To compute the value of PSNR, first, calculate the mean squared error by the following equation:
In the above equation, M represents the number of rows and N represents the columns in the images. After that the block calculates the PSNR using the following equation:
In this above
The greenish effect is caused by the color cast which was removed by using the Fusion technique. This technique can even produce better results when combined with the Adaptive color balancing technique. The next image mainly contains bluish-green effects, and the resultant image appears to correct this to some extent. However, to fully remove it, a complete proposed method needs to be implemented.
Above
Most of the researchers used images from the internet to assess the performance of their proposed techniques. However, some of them used JAMSTEC dataset to perform experiments.
Nearly every researcher used images from the internet. There are nearly 8–10 images that are used by the researcher for testing the results. We have also selected 8 images for testing purposes. The results are shown in
The proposed method also shows less value of MSE which means less error in the output image and better image quality. For internet images values came out to be 0.28 which is less than any other method.
In addition, it produces better results for EME. This indicates greater image quality improvement as compared to other methods. MOE values came out to be better than some researchers and less to only one [
To further validate the performance of the proposed framework the proposed technique was also applied to videos. Experiments are performed on the KAIKO ROV video repository and random underwater videos taken internet. Different techniques are applied to each frame to enhance the video and each performance parameter is checked accordingly to measure enhancement.
Above
References | Dataset | Performance criteria | |||
---|---|---|---|---|---|
MOE | EME | MSE | PSNR | ||
M. Hitam & A. Awalludin [ |
Images and videos from Google | 7.7752 | N/A | 0.9 | 18.89 |
Videos from YouTube and Google (“underwater videos”) | 7.02 | 12.27 | 0.28 | 31.15 |
The proposed framework is evaluated on various videos and the output is presented in
The proposed method shows less value of MSE which means less error in the output video. Our system produces less MSE compared to other methods. This method also produces fewer errors for videos.
Our method produces better results for videos, and no such preprocessing is required. MOE shows average results for images but for videos this method produces a higher value than any other method.
This method also produces a better result for EME. That indicates greater image quality improvement as compared to other methods. So overall all the values are better than the other methods.
Most of the researchers used Dark channel prior (DCP) in their approaches because it produces a higher value for Peak signal to noise ratio (PSNR) but it decreases contrast and produces a halo region. To overcome this problem CLAHE was used. After carefully considering the limitations of current methods for color correction, adaptive color correction has been presented in this literature which removes color cast and restores natural color.
Underwater images are very important for discovering new species or for photography purposes, studying fishes, and coral reefs, capturing 3D bathymetry of seafloor terrain and finding aquatic plants or saving them from extinction, etc. Underwater images suffer from low contrast, lighting conditions, underwater depth, absorption and scattering effect, low resolution, and color diminishing. This paper contains information about different underwater image enhancement techniques with their advantages and limitations. These techniques are evaluated on different performance parameters and depending upon those criteria, some techniques work best to improve visual quality and increase contrast. Only a few of the current methods handled color cast to some extent. Various researchers used Dark channel prior (DCP) in their approaches because it produces a higher value for Peak signal-to-noise ratio (PSNR) but it decreases contrast and produces a halo region. To overcome this problem CLAHE was used. After carefully considering the limitations of current methods for color correction as it overcomes the limitations of global approaches by performing local enhancement. Hence this method can be applied to a large range of images having a color cast, low contrast, and visual quality issues.
The major achievement of the proposed method is the enhanced underwater images as shown by the values of a performance parameter of underwater videos are higher for PSNR and MSE. A better value of PSNR indicates the output image has better visual quality and the mean square error of the output image is less than the previous methods. Our method improves the visual quality of underwater images better than the current methods.
The results show the potential of the proposed framework for underwater images on large datasets or videos. Our method produces higher values for PSNR, low error, and improved image quality. However, due to time constraints, the proposed framework is not developed for real-time application. There can be many future directions from this point on in this research. In the future, it can be implemented for real-time applications, and it has great potential to apply to mobile phone applications.
Researchers Supporting Project Number (RSP2022R458), King Saud University, Riyadh, Saudi Arabia.