Next Article in Journal
Zero-Day Threat Mitigation via Deep Learning in Cloud Environments
Previous Article in Journal
Unsupervised Classification and Remaining Useful Life Prediction for Turbofan Engines Using Autoencoders and Gaussian Mixture Models: A Comprehensive Framework for Predictive Maintenance
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

An Adaptive Underwater Image Enhancement Framework Combining Structural Detail Enhancement and Unsupervised Deep Fusion

Department of Computer Engineering, Faculty of Engineering, Mersin University, Mersin 33110, Turkey
*
Author to whom correspondence should be addressed.
Appl. Sci. 2025, 15(14), 7883; https://doi.org/10.3390/app15147883
Submission received: 16 June 2025 / Revised: 10 July 2025 / Accepted: 11 July 2025 / Published: 15 July 2025
(This article belongs to the Section Computing and Artificial Intelligence)

Abstract

The underwater environment severely degrades image quality by absorbing and scattering light. This causes significant challenges, including non-uniform illumination, low contrast, color distortion, and blurring. These degradations compromise the performance of critical underwater applications, including water quality monitoring, object detection, and identification. To address these issues, this study proposes a detail-oriented hybrid framework for underwater image enhancement that synergizes the strengths of traditional image processing with the powerful feature extraction capabilities of unsupervised deep learning. Our framework introduces a novel multi-scale detail enhancement unit to accentuate structural information, followed by a Latent Low-Rank Representation (LatLRR)-based simplification step. This unique combination effectively suppresses common artifacts like oversharpening, spurious edges, and noise by decomposing the image into meaningful subspaces. The principal structural features are then optimally combined with a gamma-corrected luminance channel using an unsupervised MU-Fusion network, achieving a balanced optimization of both global contrast and local details. The experimental results on the challenging Test-C60 and OceanDark datasets demonstrate that our method consistently outperforms state-of-the-art fusion-based approaches, achieving average improvements of 7.5% in UIQM, 6% in IL-NIQE, and 3% in AG. Wilcoxon signed-rank tests confirm that these performance gains are statistically significant (p < 0.01). Consequently, the proposed method significantly mitigates prevalent issues such as color aberration, detail loss, and artificial haze, which are frequently encountered in existing techniques.

1. Introduction

Underwater exploration is an important focus of research on Earth, which is approximately two-thirds water. Underwater exploration has significant contributions to many fields from various branches such as archeological remains, coral reefs, marine biology, aquatic ecosystems and unique underwater landscapes [1,2,3]. However, light reaching the underwater environment is absorbed and scattered when it interacts with water molecules and particles. Especially of note, long-wavelength lights such as red light incur the greatest losses. In contrast, green and blue lights have shorter wavelengths compared to red and can reach deep into the water [4]. Absorption and scattering of light cause many problems, such as color loss, low contrast, hazy and blurry appearance in underwater images [5]. This situation distorts the real colors and appearance of objects and the environment, culminating in perceptual weaknesses (bluish and greenish appearance), and it is a key limitation for many research areas. Therefore, in image-based underwater applications, images are first preprocessed with an image enhancement algorithm. However, owing to the complexity of the underwater environment, many advanced methods apply multiple sequential or co-occurring enhancement steps (color correction, contrast enhancement, noise reduction, haze removal and detail enhancement) and image fusion strategies to eliminate these distortions [5].
According to existing research, underwater image enhancement methods are divided into two main categories: physical models and non-physical models. In addition, deep learning models have recently been studied as a subfield of non-physical models. Physical models are based on the principle of reversing the effects of light absorption and scattering in underwater images. They focus on compensating for color loss in images and removing hazy images by estimating how much each pixel is attenuated underwater and the scattered light. For example, Image Blurriness and Light Absorption (IBLA) [6] estimates the transmission map of the light lost due to background light and absorption and scattering; Wavelength Compensation and Dehazing (WCD) [7] focuses on estimating the depth of the scene using the relationship between image blur and light absorption. Underwater Light Attenuation Prior (ULAP) [8] corrects the underwater image according to the physical model by performing fast depth estimation. Underwater Dark Channel Prior (UDCP) [9] and Underwater Total Variation (UTV) [10] adapt the dark channel assumption to the underwater environment and adapt the background light. To summarize, physically based methods are based on the high accuracy estimation of many parameters of the underwater environment such as absorption/scattering, depth map and background light. The performance of these physical model-based methods is highly dependent on the precise estimation of multiple parameters. Consequently, even minor inaccuracies in these estimations can cause the model to produce inconsistent or degraded results, particularly when applied to images from diverse underwater environments [11].
Non-physical enhancement models, instead of estimating any parameters of the underwater environment, focus on image enhancement with statistical changes, contrast enhancement, color correction and image fusion steps. For example, Relative Global Histogram Stretching (RGHS) [12], which is based on histogram equalization, focuses on contrast enhancement and color compensation by expanding the histogram of the blue and green channels, which are the dominant colors underwater, according to the global distribution. Hue-Preserving (HP) [13] aims to preserve the original color tone by applying histogram stretching to HSI/HSV color spaces. Color Balance and Fusion (CBF) [14] divides the input image into two copies, increases the contrast of one, and applies color balancing with white balance correction to the other. The two images are fused by the Laplacian pyramid to generate the final image. However, it may produce red-colored artifacts in the image. The Hybrid Fusion Method (HFM) [15] aims to comprehensively correct underwater images by blending images with multiple enhancement strategies: color and white balancing, visibility enhancement, contrast enhancement, and perceptual image fusion. The method proposed by [11], combining the improved Retinex output with Adaptive color correction with NSST-based multi-scale fusion, preserves some advantages of the physical model while offering fusion flexibility [11]. In addition, Bayesian Retinex [16] minimizes color and brightness aberrations with a statistical decomposition; Modified Color Correction + Adaptive LUT [17] limits artifacts with edge-preserving filters and LUT-based contrast enhancement. Minimal Color Loss + Locally Adaptive Contrast [18] offers a balance that minimizes color losses while adaptively increasing local contrast. In short, although methods that are not based on physical models stand out due to their fast and easy applicability, side effects such as excess color saturation or contrast, extra noise or poor generalization ability can be observed with the use of more than one method. However, acceptable results can be achieved with appropriate method selection and fine tuning [11].
Deep-learning-based methods with strong problem-solving capabilities in the field of image processing produce high-quality results by learning the mapping between corrupted images and reference images; however, they require large datasets to strengthen their generalization capabilities. A Fast Underwater Image Enhancement Generative Adversarial Network (FUnIE-GAN) [19] is a conditional convolutional GAN model. It aims to eliminate chromatic aberrations and low contrast of corrupted underwater images in real time. However, since it is a supervised learning model, it creates a large dataset with synthetic underwater images. This leads to generalization problems of the model. To address degradation diversity, Fu et al. [19] introduced SCNet to learn “water type desensitized” representations using novel normalization schemes; however, this supervised approach still relies on paired training data and its effectiveness can be limited in images suffering from extreme color casts or heavy backscatter. Although Li et al. [20] proposed Ucolor, a hybrid model using multi-color spaces, its visual results often exhibit significant color distortions and low contrast, leading to images that can appear unnatural and unrealistic. Target-Oriented Perceptual Adversarial Learning (TOPAL) [21] is a GAN-based model that focuses on improving object detection performance while enhancing underwater images. This requires multiple loss calculations, their simultaneous optimization, and specific training for each application, limiting the flexibility of the model. Unsupervised Single Underwater Image Restoration (USUIR) [22] is an unsupervised network with an encoder–decoder structure. It performs unsupervised recovery with a differentiable gradient layer and cyclic consistency using the “homology” assumption between corrupted and clean underwater images; however, if the gradient model or homology is invalid, the training becomes unstable and there is a risk of model collapse. In general, supervised deep-learning-based methods require a large number of referenced data, a need often met with distorted or synthetic data. The difficulty of obtaining referenced data in the underwater environment significantly limits the reliability and generalization ability of such methods.
In this study, we focus on the negative effects of non-physical based methods, which lead to excessive contrast or color saturation, additional noise generation, and poor generalization ability. While a dedicated processing unit strengthens the image’s structural details, we employ the unsupervised fusion network from to intelligently combine these features. This allows us to harness the powerful feature-learning capacity of deep learning, thereby achieving effective enhancement without the data dependency limitations of supervised models [23]. The main contributions of the proposed method are as follows:
I.
A two-stage Multi-Scale Detail Enhancement Unit (MSDE) is proposed to expose structural details in underwater images in a natural and distinct way.
a.
First stage: a copy of the input image is added to the original image with a detail layer created with detail maps obtained at different scales; both the edge sharpness and structural details of the image are enhanced.
b.
Second stage: the noise and artefacts that may occur as a result of the first stage are decomposed into subspaces with the Latent Low-Rank Representation (LatLRR) [24] method and unwanted effects are successfully attenuated.
II.
Adaptive gamma correction [25] is applied to another copy of the image; thus, a sensitive and dynamic brightness enhancement is provided to the brightness level in the scene.
III.
The two developed copies are combined using the MU-Fusion [26] network, an unsupervised learning method adapted to underwater images. Thus, a balanced final image covering both global contrast and local details is obtained. In addition, owing to the ability of MU-Fusion to dynamically learn information about the scene, the generalization ability of the proposed method to different underwater environments is increased.
The rest of this paper is organized as follows: In Section 2, the proposed hybrid underwater image enhancement method is presented in detail and the main steps of the method are explained. In Section 3, the experimental setup, datasets, comparison metrics, the quantitative and qualitative results, and statistical tests are comprehensively discussed. Finally, in Section 4, the general conclusions are summarized and evaluations on future work are presented.

2. Proposed Hybrid Framework for Underwater Image Enhancement

In this section, the proposed hybrid underwater image enhancement method is explained in detail. The method progressively eliminates distortions such as natural tone loss and chromatic aberrations, low contrast and loss of detail due to irregular absorption and scattering of light at different wavelengths in the underwater environment: First, the adaptive color correction algorithm [11] that is sensitive to the color distribution of the image is used; owing to traditional white balance and gray world, color correction approaches are insufficient to eliminate underwater spectral distortions and often lead to problems such as red artefacts or overcorrection [27]. Then, the YCbCr [28] color space is switched and the other steps are performed only on the Y channel. Two copies are created from the Y channel. One copy is expanded with the adaptive gamma correction proposed in [25], while the other copy enhances the structural detail and edge clarity with the detail enhancement module. Finally, when these two processed Y channels are balanced through fusion by the MU-Fusion network and integrated with the CbCr components, the final RGB image with optimized color accuracy and richness of detail is obtained, as schematically shown in Figure 1.

Proposed Multi-Scale Detail Enhancement Unit (MSDE)

The weak structural details, low contrast and blurred edges in underwater images make the detail enhancement phase a critical step in the image enhancement process. In this study, the proposed detail enhancement unit consists of two consecutive sub-steps: multi-scale detail enhancement and Latent Low-Rank Representation (LatLRR)-based structural decomposition. Using this module, the structural integrity is preserved, details are inherently emphasized, and artifacts are effectively attenuated.
Multi-Scale detail enhancement: In order to enhance the structural details, the Y channel is smoothed with Gaussian filters with different standard deviation values; then, for each scale, detail layers are created by taking the difference between the original image and its smoothed version. These detail layers contain edge and structure information of the scene at different scales. Small sigma values reveal thinner edges, while large sigma values expose wider structures. The obtained multi-scale layers are combined with certain weights to form a single detail component. This detail component is added to the input Y channel with a certain gain factor in the last stage to obtain the strengthened Y channel. Hence, this ensures that both thin edges and global details become more distinct at the same time. The mathematical modeling of the stages of the method is as follows:
The input image, represented as a normalized grayscale (Y-channel) component, is defined as follows:
I 0 ,   1 H × W
To extract structural information at multiple spatial resolutions, the image is convolved with Gaussian kernels of varying standard deviations:
G σ i = I G σ i ,                 i = 1,2 , , N
Here, G σ i denotes a 2D Gaussian filter with standard deviation ( σ i ) ,   a n d   ( ) represents the convolution operation. For each scale, a detail layer is computed by subtracting the smoothed image from the original input:
D σ i = I G σ i
These multi-scale detail maps are aggregated using scale-specific weights to form a unified detail component:
D = i = 1 N w i D σ i
Finally, the enhanced image is obtained by adding the scaled detail component to the original image:
I e n h = I + α D
where ( α ) is a gain factor controlling the strength of the detail enhancement.
Latent Low-Rank Representation (LatLRR): Although detail enhancement processes highlight structural information in the image, amplification of spurious effects such as noise and excessive emphasis may also occur. Therefore, in the proposed unit, the Latent Low-Rank Representation (LatLRR)-based structural decomposition method was preferred in order to reduce these negativities that may arise from using the detail contribution in its raw form and to extract only meaningful structural components and transfer them to the fusion stage. LatLRR enables us to represent the basic information components in the image in a more dense and separate way by decomposing the input image into three separate subspaces, namely, principal features, salient features and sparse noise [24]. In the method, only the principal features’ subspace is included in the fusion process; thus, spurious patterns or noises that may occur in the detail enhancement process are largely removed from the image. This selective decomposition approach is a critical step that supports the detail enrichment and deep-learning-based fusion process to produce more consistent results.
The LatLRR method decomposes an image matrix into three components, modeled as follows:
I = I Z + I L + E
Here, I represents the input image, I Z represents the principal features’ subspace, I L represents salient features, and E represents sparse noise. This decomposition process is performed within the framework of the following optimization problem:
m i n Z , L , E Z * + L * + λ · E 1
In this formulation, · * represents the nuclear norm, which is the sum of a matrix’s singular values and serves as a convex relaxation of the rank function. Its purpose is to encourage the primary components Z and L to be low-rank. The ·   1 denotes the l1-norm, which is applied to the error term E to model it as a sparse matrix, effectively isolating noise and gross errors. The parameter λ > 0 is a balancing coefficient that controls the trade-off between the low-rank structure and the sparse error [24].
Unsupervised Deep-Learning-Based Fusion (MU-Fusion): MU-Fusion represents one of the most critical stages of our proposed method. It is an unsupervised, deep-learning-based image fusion model that optimally combines the differently enhanced Y channels. In unsupervised fusion models, the training process continues by focusing only on the source image without transferring in-process outputs to subsequent epochs. However, in-process outputs may contain some clues for important information such as pixel intensity distribution, gradient information and structural similarity for image fusion. Owing to its memory-based architecture, MU-Fusion proposes a memory unit that allows the in-process outputs obtained in previous learning steps to affect the current output. In this way, not only the current inputs but also the in-process outputs contribute to learning; the network exhibits a more consistent and generalizable fusion performance [26]. The learning process of MUFusion is defined by a dual-component loss function that takes both content loss and memory loss into account:
L t o t a l = p 1 · L c o n t e n t + p 2 · L m e m o r y
S A = S S I M ( O , I 1 ) + S S I M ( O , I 2 ) 2
S B = S S I M ( O p r e , I 1 ) + S S I M ( O p r e , I 2 ) 2
p 1 = e S A e S A + e S B
p 2 = e S B e S A + e S B
L c o n t e n t = j = 1 2 ( L p i x e l w j · O , w j · I j   + L s s i m w j · O , w j · I j + L g r a d w j · O , w j · I j )
L m e m o r y = L p i x e l O , O p r e   + L s s i m O , O p r e + L g r a d O , O p r e
In the above formula, Lcontent measures the degree to which the obtained fusion image preserves important content details in the source images; and Lmemory ensures the consistency and continuity of the learning process with the intermediate outputs produced in the previous epochs. The parameters p 1 and p 2 control the weighting between these two types of loss and can be determined adaptively according to the image type [26]. In the proposed method, MUFusion synthesizes both global contrast and local details optimally by performing information fusion between the structural detail component has been denoised with the Latent Low-Rank Representation (LatLRR) method and the Y channel to which adaptive gamma correction has been applied. The final Y channel obtained is recombined with the Cb and Cr color components obtained in the adaptive color correction step in order to preserve color accuracy; thus, the image created in the YCbCr space is converted to the RGB color space and the final enhanced underwater image is obtained. This last step ensures that the enhancement processes only affect the structural information and do not distort the color components, thus ensuring that natural, balanced and visually rich results are obtained. This approach, which does not require a reference image, increases the generalization ability of the method and provides consistent and high-quality image enhancement under different illumination, chromatic aberration or blur conditions.

3. Results

In this section, the proposed method is evaluated quantitatively (quality metrics and statistical analyses) and qualitatively (full and cropped images, edge maps and SIFT-based) within the framework of the experimental setup, dataset and parameters.

3.1. Experimental Setup

In this study, two different real-world datasets were used to evaluate the effectiveness of the proposed hybrid underwater image enhancement method: Test-C60 and OceanDark. Both datasets consist of images reflecting difficult underwater conditions, without ground-truth references images, and containing distortions such as various color deviations and low contrast.
The Test-C60 dataset is a subset of the Underwater Image Enhancement Benchmark (UIEB) [23] dataset and consists of 60 natural underwater images. These images represent difficult scene conditions such as yellowish, greenish, and bluish tones that cannot be fully corrected by existing enhancement methods. The OceanDark [29] dataset consists of 183 low-illumination underwater images with artificial lighting collected in deep sea conditions in the Northeastern Pacific Ocean. The images were taken between 389 m and 980 m depth and include different biological species and artificial objects.
The MU-Fusion network used in the fusion phase of the proposed method was trained on 100 randomly selected raw underwater images from the UIEB dataset. During the training process, the learning rate of the model was determined as 10−4, and training was performed for a total of 10 epochs. To properly configure the content loss, we set the two key parameters that control the activity level map generation: the feature map depth was set to 5 to leverage deep semantic information for identifying salient regions, while the local averaging window size was set to 3 to ensure the preservation of fine details. No special data augmentation method was used in the training process, and raw images were evaluated directly. Dataset distributions for MUFusion model training are given below in Table 1.
Experimental studies were carried out on a computer with Intel i5-13600K processor, 32 GB RAM and NVIDIA RTX 4070 Ti GPU hardware. Various reference-free image quality metrics (UIQM, UICM, AG, IE and IL-NIQE) are used to evaluate the quantitative success of the proposed method, and Scale-Invariant Feature Transform (SIFT) [30]-based matching evaluations are performed for structural similarity analysis. The statistical significance of the obtained results is analyzed by the Wilcoxon signed-rank test.

3.2. Quantitative Evaluation

In order to comprehensively evaluate the success of the proposed method in this study, both perceptually and structurally, various reference-free image quality metrics were used. Each of these metrics allows us to evaluate the performance of the methods by measuring different aspects of image quality. Integrated Local Natural Image Quality Evaluator (IL-NIQE) [31] is a completely blind quality indicator and measures the deviation from natural scene statistics; lower IL-NIQE values indicate that noise, blur and color errors are reduced, and the perceived quality is increased. Average Gradient (AG) [20] measures the average of spatial gradient sizes in the image; high AG values indicate that detail richness and edge sharpness are improved. However, when the AG values increase excessively, this increase can sometimes be caused by unwanted noise; so, it should be interpreted together with IL-NIQE. Underwater Image Quality Measure (UIQM) [6] is a composite metric based on the human-eye model developed for underwater images and is calculated as a weighted sum of elements such as the underwater image colorfulness measure (UICM), the underwater image contrast measure (UIConM) and the underwater image sharpness measure (UISM); therefore, higher UIQM values indicate increased overall enhancement quality. UICM [6] specifically measures color shifts and color saturation; higher UICM values indicate that color aberrations are eliminated and color vibrancy is increased. Information Entropy (IE) [32] measures the information content of the image and the uncertainty in tone distribution; higher IE generally means wider dynamic range and increased information.
Metric comparison: Table 2 presents the average values of IE, IL-NIQE, AG, and UIQM and its sub-component UICM metrics calculated for 60 underwater images in the Test-C60 dataset, and Table 3 presents the average values of 184 underwater images in the OceanDark dataset. These tables clearly demonstrate the performance superiority of the proposed method over 14 other underwater image enhancement methods in terms of these five metrics.
When the scores presented in Table 2 are examined, it is seen that our proposed method produces the results closest to natural scene statistics by obtaining the lowest value (25.46) in IL-NIQE and records the highest value (0.997) in UIQM, proving that it improves color, contrast and sharpness in a balanced manner. The detail preservation ability of the method is confirmed by the best score of 51.60 reached in the average gradient (AG) metric; and in UICM, representing color accuracy, the second-best value (11.05) is achieved right after the aggressive color compensation of HFM. Although it remains in the third place in terms of entropy (15.32), the low value of IL-NIQE and the high value of AG show that this information is due to a balanced contrast increase and does not consist of excessive noise. Other compared methods either produced blur while increasing color saturation (HFM) or were insufficient in providing detail [11]. The results show that the proposed method achieves consistent success in both objective and perceptual metrics by finding the optimum point in the color-detail trade-off on the Test-C60 dataset.
The average scores calculated on 183 low-light and dense scattering images in the OceanDark dataset in Table 3 show that the proposed method outperforms its competitors by 2.6–19 points, achieving the lowest value of 23.06 in IL-NIQE. It surpasses the second-ranked [11] by ≈13% and HFM by ≈23% in UIQM with a score of 1.1182, indicating that it improves color, contrast and sharpness in a balanced manner. Although it only lags behind HFM’s aggressive color compensation (16.58) with a value of 15.50 in UICM, indicating color accuracy, its superiority over IL-NIQE reveals that HFM achieves this gain at the expense of noise. It suppresses blur most effectively in the AG metric, indicating detail preservation, by 48.99, leaving its closest competitor with a difference of about 6 points. It ranks high with an entropy value of 16.36, confirming the increased detail and color volume, while avoiding high IL-NIQE scores indicating an increase in spurious information due to noise. These results show that the proposed method provides a strong and consistent performance in both objective and perceptual metrics, preserving the color–detail balance even in the challenging low-illumination conditions of OceanDark.
Statistical significance analysis: In order to evaluate whether the performance differences of the proposed method against different image enhancement methods are not only numerically but also statistically significant, the Wilcoxon signed-rank test was applied. This test is especially suitable for nonparametric comparisons between dependent samples and measures the significance of the differences between the metric scores of each method pair. In the analyses, the p-values of the comparisons made between the proposed method and other methods according to the IE, IL-NIQE, AG, UIQM and UICM metrics were calculated and p < 0.01 was regarded as the significance level. Table 4 presents the statistical comparisons made between the proposed method and other methods on the Test-C60 dataset, and Table 4 presents the statistical comparisons made between the proposed method and other methods on the OceanDark dataset.
As shown in Table 4, it is seen that the proposed method is significant superior in comparison with the majority of the compared methods in all metrics of IE, IL-NIQE, AG, UIQM and UICM. In particular, significant differences at the level of p < 0.01 were detected in all metrics in comparisons made with methods such as HFM, ULAP, UDCP and FUnIEGAN. Although there are partial exceptions in some comparisons made with UT, TOPAL, RGHS and Lin et al.’s methods, especially in metrics such as AG and UICM, the general trend shows that the proposed method provides a statistically strong superiority.
Similar to Table 4, Table 5 shows that the proposed method has significant superiority in all metrics with extremely low p-values. Especially in the OceanDark dataset, which includes low-illumination and hazy scenes, statistical significance was achieved at the p < 0.01 level against methods such as WCD, UTV, ULAP, UDCP, RGHS, HP and FUnIEGAN in all metrics. In comparison with methods such as UT, TOPAL and Lin et al.’s, significant differences were found in the IE, UIQM and IL-NIQE metrics, while only some AG and UICM comparisons had borderline significance. As a result of the numerical analyses performed, it is proven that the proposed method provides consistent superiority over existing methods in terms of perceptual quality, structural detail preservation and color accuracy in both Test-C60 and OceanDark datasets. Especially the best scores obtained in critical metrics such as IL-NIQE, UIQM and AG show that the proposed approach can increase contrast and detail without creating noise and produce results compatible with natural scene statistics by minimizing color distortions. In addition, with the Wilcoxon signed-rank tests, a significant superiority at p < 0.01 level was achieved over the existing methods compared in IE, IL-NIQE, AG, UIQM and UICM metrics, thus statistically strongly supporting that the proposed method offers a reliable and effective solution in the field of underwater image enhancement.

3.3. Qualitative Evaluation

In order to evaluate the performance of the proposed method based on human visual perception and its qualitative enhancement ability, visual comparisons were made on examples representing different underwater image characteristics. Due to the refraction, absorption and scattering of light in the underwater environment, images are often distorted in hazy, greenish, bluish or yellowish tones. Therefore, scenes representing different types of distortion were selected from both Test-C60 and OceanDark datasets. Figure 2 shows three different distortion examples from the Test-C60 dataset: the first image is distorted in yellowish, the second in bluish and the third in greenish tones. These scenes illustrate common distortions specific to the underwater environment such as chromatic aberration and loss of contrast. Figure 3 presents scenes with low-illumination conditions and heavy haze effects from the OceanDark dataset. In both figures, the outputs of twelve successful underwater image enhancement methods are shown comparatively with the proposed method.
In the comparison in Figure 2, Lin et al.’s, HFM and the proposed method are seen to be superior to other methods in color accuracy capture and contrast enhancement. Among these three methods, HFM is slightly behind in correcting yellowish tones and balancing contrast in bluish images; Lin et al.’s method produced better results in removing color tones and increasing contrast compared to HFM. The proposed method, on the other hand, exhibited the most successful performance in terms of both color accuracy and detail enhancement in all tones; it made background details more distinct especially in yellowish images, and stood out in both color and detail richness in bluish and greenish images.
Similarly, in Figure 3, Lin et al.’s, HFM and the proposed method stand out as the three most successful methods. The proposed method provided the best balance in haze removal and contrast enhancement, providing perceptually superior results compared to Lin et al.’s and HFM. In general, the evaluations of Figure 2 and Figure 3 clearly show that the proposed method provides the best performance in terms of color balance, contrast enhancement, detail enhancement and perceptual improvement. The obtained quality metrics and Wilcoxon analysis results also confirm this success.

3.4. Quantitative SIFT Analysis and Statistical Validation

With the objective of evaluating the success of the proposed method in structural preservation, the Scale-Invariant Feature Transform (SIFT) algorithm, which is insensitive to scale and orientation changes and provides rotation-invariant feature extraction, was used on all methods. SIFT objectively measures the structural similarity of the enhanced images to the original raw images by detecting characteristic edges, corners and structure points in the images. In this direction, the number of matching key points detected between the enhanced images and the original images for each method was calculated and the average scores are presented in Table 6. Furthermore, the statistical significance of the structural differences between the methods was analyzed with the Wilcoxon signed-rank test, and the results are given in Table 7. The high number of SIFT matches indicates the success of the method in preserving details and improving structural integrity; however, the reliability of this increase should be supported by evaluating it together with quality metrics such as IL-NIQE or AG.
The proposed method outperforms all competitors with an average of 2923 matches on the Test-C60 dataset, increasing the structural detail saliency by approximately 3.9 times compared to the input image. It also outperforms its closest follower, Lin et al.’s method, by 7% and classical methods such as IBLA and HP by 30–50%. On the OceanDark dataset, a 46% increase is achieved compared to the baseline input with an average of 6122 matches. This high match rate leaves Lin et al.’s, HFM, and other color-based enhancement methods behind; it ranks second only to UTV’s extremely high number of matches (8470). However, when this extraordinary increase in UTV on OceanDark is evaluated together with the serious deterioration in the IL-NIQE score (38.82), it is understood that a significant portion of the recorded matches are due to noise.
The SIFT-based Wilcoxon signed-rank test results presented in Table 7 show that the proposed method provides statistically significant superiority in terms of preserving structural details in both the Test-C60 and OceanDark datasets. When compared against all methods in Test-C60, significant differences were obtained at the level of p < 0.01, and no significance was observed only against Lin et al.’s method (p = 0.388). In the OceanDark dataset, extremely strong differences were detected, especially against methods such as UT, FUnIEGAN and HFM at the level of p < 10−30. The SIFT-based analyses performed revealed that the proposed method preserved structural details with the highest accuracy in both the Test-C60 and OceanDark datasets. Although no statistically significant difference was obtained in terms of SIFT metric in the comparison with the [11] method, the fact that this method has lower AG and higher IL-NIQE scores compared to the proposed method indicates that some of the obtained matches may be due to noise. In addition, the SIFT match images of the two leading methods, Lin et al.’s and HFM, and our proposed method are shown in Figure 4. In the figure, it is clearly observed that the proposed method preserves the detail integrity qualitatively and superiorly.

3.5. Comparative Analysis of Top-Performing Methods

In this section, a detailed comparison of the proposed method with the [11] and HFM methods on randomly selected examples from the datasets is presented. In Figure 5, the saliency maps which illustrate the success of each method in highlighting important de-tails in the image are presented; in Figure 6, the zoomed-in views of the selected region sections are shown to compare the detail enhancement performance; in Figure 7, the edge maps created to objectively evaluate the edge enhancement performance and the relevant edge intensity score in the upper left corner are presented; and in Figure 8 presents the comparison of the reference-free quality metric scores of the images in Figure 2. The current assess-ments aim to reveal the differences between the methods in terms of detail preservation and structural integrity at both global and local levels.

4. Conclusions

The method proposed in this study offers a hybrid underwater image enhancement approach focused on a single scene, which does not require a large amount of dataset and reference images. The proposed detail enhancement unit, as one of the basic building blocks of the method, combines multi-scale detail highlighting with structural decomposition, enabling both the highlighting of real details and the suppression of noise and artifacts. This unit produces a simplified content that preserves structural information and directly contributes to the final fusion quality. In addition, the method, which combines image-processing techniques with the powerful representation capacity of unsupervised deep learning, minimizes fundamental distortions such as chromatic aberration, contrast loss and structural blurring caused by the intense absorption and scattering of light in the underwater environment. In order to demonstrate the effectiveness of the method, experiments conducted on images obtained from different underwater environments such as hazy, bluish, greenish and yellowish quantitatively demonstrate that the proposed method outperforms the current state-of-the-art methods in metrics such as UIQM, UICM, IL-NIQE, IE and AG. In addition, in the qualitative evaluation, it is observed that it visually eliminates color distortions, clarifies the details and produces results more suitable for the human visual system. The proposed method also achieves successful results in tests performed with Scale-Invariant Feature Transform (SIFT) and produces results comparable to those of outstanding approaches such as Lin et al.’s and HFM. The statistical significance of the proposed method’s performance was rigorously evaluated using the Wilcoxon signed-rank test on all reference-free quality metrics. This analysis substantiates that the observed superiority is not a random occurrence, thereby statistically validating the effectiveness of our approach. As a result, the proposed hybrid method makes significant progress in the field of underwater image enhancement and contributes to obtaining higher quality and detailed underwater images in various applications.
While the proposed framework has demonstrated strong performance in enhancing underwater images, it is important to acknowledge its inherent limitations, which also pave the way for future research. The primary limitation is the trade-off between detail enhancement and potential noise amplification. Our method’s design prioritizes the recovery of fine textures and structural details; a consequence of this is that in some images, latent noise can become more prominent compared to methods that employ heavier smoothing. A second practical limitation is the computational cost. As the primary focus of this study was to validate the model’s effectiveness, the current implementation has not been optimized for runtime speed, which may limit its use in real-time applications.
Future work will focus on two key areas: (1) integrating more sophisticated noise-aware modules that can better distinguish between fine textures and unwanted noise, and (2) optimizing the framework for computational efficiency to facilitate its deployment on resource-constrained platforms. This approach can both reduce processing time and facilitate the integration of the method into embedded systems and real-time underwater imaging applications.

Author Contributions

Conceptualization, S.K. and E.A.; Methodology, S.K. and E.A.; Software, S.K.; Validation, S.K.; Formal Analysis, S.K.; Investigation, S.K.; Resources, S.K.; Data Curation, S.K.; Writing—Original Draft Preparation, S.K.; Writing—Review and Editing, S.K. and E.A.; Visualization, S.K.; Supervision, E.A.; Project Administration, E.A.; Funding Acquisition, S.K. All authors have read and agreed to the published version of the manuscript.

Funding

This study is supported by the Mersin University Scientific Research Projects Coordination Unit. Project Number: 2023-2-TP3-4925.

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

The data presented in this study are available on request from the corresponding author.

Conflicts of Interest

The authors declare no conflicts of interest. The funders had no role in the design of this study; in the collection, analyses, or interpretation of data; in the writing of the manuscript; or in the decision to publish the results.

References

  1. Shortis, M.; Abdo, E.H.D. A Review of Underwater Stereo-Image Measurement for Marine Biology and Ecology Applications. In Oceanography and Marine Biology; CRC Press: Boca Raton, FL, USA, 2016; pp. 269–304. [Google Scholar]
  2. Yan, X.; Wang, G.; Jiang, G.; Wang, Y.; Mi, Z.; Fu, X. A Natural-Based Fusion Strategy for Underwater Image Enhancement. Multimed. Tools Appl. 2022, 81, 30051–30068. [Google Scholar] [CrossRef]
  3. Kocak, D.M.; Dalgleish, F.R.; Caimi, F.M.; Schechner, Y.Y. A Focus on Recent Developments and Trends in Underwater Imaging. Mar. Technol. Soc. J. 2008, 42, 52–67. [Google Scholar] [CrossRef]
  4. Yang, M.; Hu, J.; Li, C.; Rohde, G.; Du, Y.; Hu, K. An In-Depth Survey of Underwater Image Enhancement and Restoration. IEEE Access 2019, 7, 123638–123657. [Google Scholar] [CrossRef]
  5. Li, C.; Guo, J.; Guo, C. Emerging from Water: Underwater Image Color Correction Based on Weakly Supervised Color Transfer. IEEE Signal Process. Lett. 2018, 25, 323–327. [Google Scholar] [CrossRef]
  6. Panetta, K.; Gao, C.; Agaian, S. Human-Visual-System-Inspired Underwater Image Quality Measures. IEEE J. Ocean. Eng. 2015, 41, 541–551. [Google Scholar] [CrossRef]
  7. Peng, Y.-T.; Cosman, P.C. Underwater Image Restoration Based on Image Blurriness and Light Absorption. IEEE Trans. Image Process. 2017, 26, 1579–1594. [Google Scholar] [CrossRef]
  8. Song, W.; Wang, Y.; Huang, D.; Tjondronegoro, D. A Rapid Scene Depth Estimation Model Based on Underwater Light Attenuation Prior for Underwater Image Restoration. In Proceedings of the Advances in Multimedia Information Processing—PCM 2018: 19th Pacific-Rim Conference on Multimedia, Hefei, China, 21–22 September 2018; Proceedings, Part I 19. Springer: Cham, Switzerland, 2018; pp. 678–688. [Google Scholar]
  9. Song, W.; Wang, Y.; Huang, D.; Liotta, A.; Perra, C. Enhancement of Underwater Images with Statistical Model of Background Light and Optimization of Transmission Map. IEEE Trans. Broadcast. 2020, 66, 153–169. [Google Scholar] [CrossRef]
  10. Hou, G.; Li, J.; Wang, G.; Yang, H.; Huang, B.; Pan, Z. A Novel Dark Channel Prior Guided Variational Framework for Underwater Image Restoration. J. Vis. Commun. Image Represent. 2020, 66, 102732. [Google Scholar] [CrossRef]
  11. Lin, S.; Li, Z.; Zheng, F.; Zhao, Q.; Li, S. Underwater Image Enhancement Based on Adaptive Color Correction and Improved Retinex Algorithm. IEEE Access 2023, 11, 27620–27630. [Google Scholar] [CrossRef]
  12. Huang, D.; Wang, Y.; Song, W.; Sequeira, J.; Mavromatis, S. Shallow-Water Image Enhancement Using Relative Global Histogram Stretching Based on Adaptive Parameter Acquisition. In Proceedings of the MultiMedia Modeling: 24th International Conference, MMM 2018, Bangkok, Thailand, 5–7 February 2018; Proceedings, Part I 24. Springer: Cham, Switzerland, 2018; pp. 453–465. [Google Scholar]
  13. Hou, G.; Pan, Z.; Huang, B.; Wang, G.; Luan, X. Hue Preserving-based Approach for Underwater Colour Image Enhancement. IET Image Process. 2018, 12, 292–298. [Google Scholar] [CrossRef]
  14. Ancuti, C.O.; Ancuti, C.; De Vleeschouwer, C.; Bekaert, P. Color Balance and Fusion for Underwater Image Enhancement. IEEE Trans. Image Process. 2017, 27, 379–393. [Google Scholar] [CrossRef] [PubMed]
  15. An, S.; Xu, L.; Senior Member, I.; Deng, Z.; Zhang, H. HFM: A Hybrid Fusion Method for Underwater Image Enhancement. Eng. Appl. Artif. Intell. 2024, 127, 107219. [Google Scholar] [CrossRef]
  16. Zhuang, P.; Li, C.; Wu, J. Bayesian Retinex Underwater Image Enhancement. Eng. Appl. Artif. Intell. 2021, 101, 104171. [Google Scholar] [CrossRef]
  17. Muniraj, M.; Dhandapani, V. Underwater Image Enhancement by Modified Color Correction and Adaptive Look-Up-Table with Edge-Preserving Filter. Signal Process. Image Commun. 2023, 113, 116939. [Google Scholar] [CrossRef]
  18. Zhang, W.; Zhuang, P.; Sun, H.-H.; Li, G.; Kwong, S.; Li, C. Underwater Image Enhancement via Minimal Color Loss and Locally Adaptive Contrast Enhancement. IEEE Trans. Image Process. 2022, 31, 3997–4010. [Google Scholar] [CrossRef]
  19. Fu, Z.; Lin, X.; Wang, W.; Huang, Y.; Ding, X. Underwater Image Enhancement via Learning Water Type Desensitized Representations. In Proceedings of the ICASSP 2022-2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Singapore, 23–27 May 2022; IEEE: Piscataway, NJ, USA, 2022; pp. 2764–2768. [Google Scholar]
  20. Li, C.; Anwar, S.; Hou, J.; Cong, R.; Guo, C.; Ren, W. Underwater Image Enhancement via Medium Transmission-Guided Multi-Color Space Embedding. IEEE Trans. Image Process. 2021, 30, 4985–5000. [Google Scholar] [CrossRef]
  21. Jiang, Z.; Li, Z.; Yang, S.; Fan, X.; Liu, R. Target Oriented Perceptual Adversarial Fusion Network for Underwater Image Enhancement. IEEE Trans. Circuits Syst. Video Technol. 2022, 32, 6584–6598. [Google Scholar] [CrossRef]
  22. Fu, Z.; Lin, H.; Yang, Y.; Chai, S.; Sun, L.; Huang, Y.; Ding, X. Unsupervised Underwater Image Restoration: From a Homology Perspective. Proc. AAAI Conf. Artif. Intell. 2022, 36, 643–651. [Google Scholar] [CrossRef]
  23. Li, C.; Guo, C.; Ren, W.; Cong, R.; Hou, J.; Kwong, S.; Tao, D. An Underwater Image Enhancement Benchmark Dataset and Beyond. IEEE Trans. Image Process. 2019, 29, 4376–4389. [Google Scholar] [CrossRef]
  24. Liu, G.; Yan, S. Latent Low-Rank Representation for Subspace Segmentation and Feature Extraction. In Proceedings of the 2011 International Conference on Computer Vision, Barcelona, Spain, 6–13 November 2011; IEEE: Piscataway, NJ, USA, 2011; pp. 1615–1622. [Google Scholar]
  25. Rahman, S.; Rahman, M.M.; Abdullah-Al-Wadud, M.; Al-Quaderi, G.D.; Shoyaib, M. An Adaptive Gamma Correction for Image Enhancement. EURASIP J. Image Video Process. 2016, 2016, 35. [Google Scholar] [CrossRef]
  26. Cheng, C.; Xu, T.; Wu, X.-J. MUFusion: A General Unsupervised Image Fusion Network Based on Memory Unit. Inf. Fusion 2023, 92, 80–92. [Google Scholar] [CrossRef]
  27. Ebner, M. Color Constancy. In Computer Vision: A Reference Guide; Springer International Publishing: Cham, Switzerland, 2021; pp. 168–175. [Google Scholar]
  28. Gonzalez, R.C. Digital Image Processing; Pearson Education India: Chennai, India, 2009; ISBN 8131726959. [Google Scholar]
  29. Marques, T.P.; Albu, A.B. L2uwe: A Framework for the Efficient Enhancement of Low-Light Underwater Images Using Local Contrast and Multi-Scale Fusion. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, Seattle, WA, USA, 14–19 June 2020; pp. 538–539. [Google Scholar]
  30. Lowe, D.G. Distinctive Image Features from Scale-Invariant Keypoints. Int. J. Comput. Vis. 2004, 60, 91–110. [Google Scholar] [CrossRef]
  31. Zhang, L.; Zhang, L.; Bovik, A.C. A Feature-Enriched Completely Blind Image Quality Evaluator. IEEE Trans. Image Process. 2015, 24, 2579–2591. [Google Scholar] [CrossRef] [PubMed]
  32. Núñez, J.A.; Cincotta, P.M.; Wachlin, F.C. Information Entropy: An Indicator of Chaos. In Proceedings of the Chaos in Gravitational N-Body Systems, Proceedings of a Workshop held at La Plata (Argentina), La Plata, Argentina, 31 July–3 August 1995; Springer: Cham, Switzerland, 1996; pp. 43–53. [Google Scholar]
Figure 1. Flow chart of the proposed hybrid framework.
Figure 1. Flow chart of the proposed hybrid framework.
Applsci 15 07883 g001
Figure 2. Visual comparison of enhancement results for selected underwater images from the Test-C60 dataset. Each row corresponds to a different input image type (e.g., greenish, bluish, yellowish), while each column shows the result obtained by a different enhancement method, including the proposed method [11].
Figure 2. Visual comparison of enhancement results for selected underwater images from the Test-C60 dataset. Each row corresponds to a different input image type (e.g., greenish, bluish, yellowish), while each column shows the result obtained by a different enhancement method, including the proposed method [11].
Applsci 15 07883 g002
Figure 3. Visual comparison of enhancement results for selected underwater images from the OceanDark dataset. Each row corresponds to a different input image type (e.g., hazy, low-contrast), while each column shows the result obtained by a different enhancement method, including the proposed method [11].
Figure 3. Visual comparison of enhancement results for selected underwater images from the OceanDark dataset. Each row corresponds to a different input image type (e.g., hazy, low-contrast), while each column shows the result obtained by a different enhancement method, including the proposed method [11].
Applsci 15 07883 g003aApplsci 15 07883 g003b
Figure 4. Visual comparison of SIFT keypoint-matching results. The proposed method detects and preserves the most distinctive feature points, resulting in more accurate and denser matching compared to competing methods [11].
Figure 4. Visual comparison of SIFT keypoint-matching results. The proposed method detects and preserves the most distinctive feature points, resulting in more accurate and denser matching compared to competing methods [11].
Applsci 15 07883 g004
Figure 5. Saliency maps obtained on sample images randomly selected from different datasets. The proposed method makes the details more distinct by highlighting the regions of interest more sharply and clearly compared to other methods [11].
Figure 5. Saliency maps obtained on sample images randomly selected from different datasets. The proposed method makes the details more distinct by highlighting the regions of interest more sharply and clearly compared to other methods [11].
Applsci 15 07883 g005aApplsci 15 07883 g005b
Figure 6. Cropped region comparison of enhanced underwater images. Close-up views highlight the proposed method’s superiority in preserving fine details and suppressing artifacts in structurally complex regions [11].
Figure 6. Cropped region comparison of enhanced underwater images. Close-up views highlight the proposed method’s superiority in preserving fine details and suppressing artifacts in structurally complex regions [11].
Applsci 15 07883 g006
Figure 7. Edge maps of enhanced underwater images generated by different methods. The proposed method preserves structural integrity and enhances fine edges more effectively [11].
Figure 7. Edge maps of enhanced underwater images generated by different methods. The proposed method preserves structural integrity and enhances fine edges more effectively [11].
Applsci 15 07883 g007aApplsci 15 07883 g007b
Figure 8. Quantitative comparison of enhanced images in terms of IL-NIQE (↓), AG (↑), IE (↑), and UIQM (↑) for three different rows presented in Figure 2. The proposed method consistently achieves lower IL-NIQE scores and higher AG and UIQM scores, indicating superior perceptual quality and edge sharpness across all cases [11].
Figure 8. Quantitative comparison of enhanced images in terms of IL-NIQE (↓), AG (↑), IE (↑), and UIQM (↑) for three different rows presented in Figure 2. The proposed method consistently achieves lower IL-NIQE scores and higher AG and UIQM scores, indicating superior perceptual quality and edge sharpness across all cases [11].
Applsci 15 07883 g008
Table 1. Dataset distributions for MUFusion model training.
Table 1. Dataset distributions for MUFusion model training.
DatasetNumber of ImagesPurpose of UseExplanation
UIEB(Training Subset)100TrainingDivided into 19,375 patches of size 128 × 128
UIEB—TestC6060TestingNot used during training
OceanDark183TestingNot used during training
Table 2. Quantitative comparison of average metric scores on the Test-C60 dataset using reference-free quality metrics.
Table 2. Quantitative comparison of average metric scores on the Test-C60 dataset using reference-free quality metrics.
Test-C60
IE ↑UIQM ↑UICM ↑ILNIQE ↓AG ↓
IBLA14.39130.51445.838735.613630.8885
CBF14.76930.60449.944929.700130.5803
WCD12.96840.30327.192340.586124.1967
ULAP12.63570.36575.357641.648224.6411
RGHS14.66620.47695.629932.393831.7535
HP15.19230.47377.258733.410134.9702
UTV11.23130.20092.462633.427822.2727
UDCP12.13850.09532.635241.578424.5690
FUnIEGAN14.23080.38246.795750.677545.0743
TOPAL14.45510.27775.952846.754747.1259
USUIR14.91810.670410.21929.672830.2130
UT14.59390.38827.953945.768048.0182
[11]15.45770.927410.66227.076150.0053
HFM15.60020.869714.71829.388432.7669
Proposed15.32210.997111.05425.462251.60
Arrows next to the metric values indicate whether a higher (↑) or lower (↓) score represents better performance for that specific metric.
Table 3. Quantitative comparison of average metric scores on the OceanDark dataset using reference-free quality metrics.
Table 3. Quantitative comparison of average metric scores on the OceanDark dataset using reference-free quality metrics.
OceanDark
IE ↑UIQM ↑UICM ↑ILNIQE ↓AG ↑
IBLA14.28140.770611.835530.640228.7766
CBF14.99590.649411.0727.352224.50
WCD14.24660.28303.039928.440928.7920
ULAP13.71310.627610.086534.214225.0536
RGHS15.56450.50994.256928.797225.5566
HP15.53110.53727.281833.679323.8611
UTV12.97530.53723.521838.820031.2970
UDCP13.25670.16021.467332.206525.3601
FUnIEGAN15.3540.30116.522350.343539.8643
TOPAL15.61650.435811.47242.024143.0719
USUIR15.55210.50584.082027.329325.2927
UT15.51590.464711.285842.967842.1335
[11]15.56100.989913.799525.838440.9591
HFM15.94090.911916.582725.620229.2301
Our15.36251.118215.497823.061048.9899
Arrows next to the metric values indicate whether a higher (↑) or lower (↓) score represents better performance for that specific metric.
Table 4. Wilcoxon signed-rank test results on the Test-C60 dataset for five reference-free quality metrics.
Table 4. Wilcoxon signed-rank test results on the Test-C60 dataset for five reference-free quality metrics.
Proposed vs.IEIL-NIQEAGUIQMUICM
R+R−pR+R−pR+R−pR+R−pR+R−p
WCD183001.629 × 10−11518252.097 × 10−11182911.714 × 10−11183001.629 × 10−1113814490.000602
UTV183001.629 × 10−11718232.319 × 10−11183001.629 × 10−11182821.803 × 10−1116771532.028 × 10−8
UT1759715.191 × 10−10018301.629 × 10−1111237070.12571182462.205 × 10−1113005300.00459
USUIR15133171.071 × 10−54417861.436 × 10−10183001.629 × 10−111801296.920 × 10−1110967340.18271
ULAP182911.714 × 10−11018301.629 × 10−11183001.629 × 10−11183001.629 × 10−1116292011.470 × 10−7
UDCP183001.629 × 10−11018301.629 × 10−11183001.629 × 10−11183001.629 × 10−1117231072.711 × 10−9
TOPAL1740901.252 × 10−9118291.714 × 10−1111606700.07129182911.714 × 10−1115023281.551 × 10−5
RGHS15862447.826 × 10−71418163.293 × 10−11183001.629 × 10−11182552.097 × 10−1116062243.640 × 10−7
[11]46813620.0009929915315.767 × 10−611996310.0365183001.629 × 10−1110527780.31319
IBLA16351951.155 × 10−71518153.462 × 10−11183001.629 × 10−1114843462.804 × 10−515712591.370 × 10−6
HP12146160.0278118291.714 × 10−11183001.629 × 10−111813173.824 × 10−1114353950.000129
HFM35014803.192 × 10−510217282.164 × 10−9183001.629 × 10−11183001.629 × 10−1125715731.272 × 10−6
FUnIEGAN1797338.41807 × 10−11018301.62956 × 10−1112325980.01961520114134170.00024629913884420.000497613
CBF15902406.727 × 10−75717732.679 × 10−10183001.629 × 10−111804265.97087 × 10−1111456850.090423362
Table 5. Wilcoxon signed-rank test results on the OceanDark dataset for five reference-free quality metrics.
Table 5. Wilcoxon signed-rank test results on the OceanDark dataset for five reference-free quality metrics.
Proposed vs.IEIL-NIQEAGUIQMUICM
R+R−pR+R−pR+R−pR+R−pR+R−p
WCD16,83608.806 × 10−3247216,3641.685 × 10−2816,83608.806 × 10−3216,83608.80695 × 10−3216,7051317.48905 × 10−31
UTV16,793431.784 × 10−311016,8261.038 × 10−3116,83249.406 × 10−3216,83608.80695 × 10−3216,740964.24099 × 10−31
UT334713,4891.583 × 10−12016,8368.806 × 10−3212,43644002.149 × 10−816,83608.80695 × 10−3214,10227342.35187 × 10−15
USUIR535511,4811.967 × 10−58716,7493.662 × 10−3116,83608.806 × 10−3216,83608.80695 × 10−3216,778582.28128 × 10−31
ULAP16,808281.395 × 10−31116,8358.953 × 10−3216,83608.806 × 10−3216,83608.80695 × 10−3215,55412822.65895 × 10−23
UDCP16,83608.806 × 10−32916,8271.021 × 10−3116,83608.806 × 10−3216,83608.80695 × 10−3216,5682686.78004 × 10−30
TOPAL380013,0361.22 × 10−10016,8368.806 × 10−3211,61552218.376 × 10−616,83608.80695 × 10−3214,94318939.61308 × 10−20
RGHS483412,0025.893 × 10−71316,8231.090 × 10−3116,83608.806 × 10−3216,83608.80695 × 10−3216,7131236.57763 × 10−31
[11]431112,5251.043 × 10−875116,0851.199 × 10−2614,41724196.26087 × 10−1716,4703663.20643 × 10−2914,66221743.27127 × 10−18
IBLA13,17036663.534 × 10−114616,7901.874 × 10−3116,83608.80695 × 10−3216,5343021.16481 × 10−2914,57322639.68665 × 10−18
HP395712,8795.072 × 10−10016,8368.806 × 10−3216,83608.80695 × 10−3216,83608.80695 × 10−3216,5313051.22164 × 10−29
HFM49516,3412.40996 × 10−28178515,0512.380 × 10−2016,83608.80695 × 10−3216,3215153.2853 × 10−28552211,3145.44014 × 10−5
FUnIEGAN628110,5550.0029016,8368.806 × 10−3212,68841482.67044 × 10−916,83608.80695 × 10−3216,6981388.38872 × 10−31
CBF12,41844182.483 × 10−810516,7314.909 × 10−3116,83608.80695 × 10−3216,789471.90546 × 10−3114,02128155.79458 × 10−15
Table 6. Total number of SIFT keypoint matches between original and enhanced images for each method on the Test-C60 and OceanDark datasets.
Table 6. Total number of SIFT keypoint matches between original and enhanced images for each method on the Test-C60 and OceanDark datasets.
Test-C60OceanDark
SIFTSIFT
Input7484180
IBLA19645337
CBF16914458
WCD9775778
ULAP12565837
RGHS19925226
HP19574582
UTV10488470
UDCP10465553
FUnIEGAN152588
TOPAL217789
USUIR17534811
UT200525
[11]27395989
HFM21655677
Proposed29236122
Table 7. SIFT-based Wilcoxon significance test results for Test-C60 and OceanDark datasets.
Table 7. SIFT-based Wilcoxon significance test results for Test-C60 and OceanDark datasets.
Proposed vs.Test-C60OceanDark
R+R−pR+R−p
WCD195307.54443 × 10−1210,57255380.000286733
UTV1827.5125.52.41749 × 10−9407811,8531.63027 × 10−8
UT195128.3325 × 10−1216,83608.71027 × 10−32
USUIR1723472.50036 × 10−1014,094.51836.55.2497 × 10−19
ULAP1790401.17628 × 10−1010,123.54927.58.15131 × 10−5
UDCP1943101.22579 × 10−11985657200.002242207
TOPAL1950.52.58.54366 × 10−1216,65301.26866 × 10−31
RGHS1582.5187.51.39138 × 10−712,17132292.6401 × 10−11
Input195307.5411 × 10−1214,05611693.459 × 10−22
[11]1099.5853.50.3882448159243.56156.50.021385042
IBLA1668.5222.52.05359 × 10−711,018.54381.57.5607 × 10−7
HP1651.5178.55.86926 × 10−814,028.51902.51.25845 × 10−18
HFM15861841.20452 × 10−710,432.55320.50.000179125
FUnIEGAN1951.51.58.12526 × 10−1216,83608.72993 × 10−32
CBF1645.565.59.5145 × 10−1014,208.51901.57.57233 × 10−19
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Kahveci, S.; Avaroğlu, E. An Adaptive Underwater Image Enhancement Framework Combining Structural Detail Enhancement and Unsupervised Deep Fusion. Appl. Sci. 2025, 15, 7883. https://doi.org/10.3390/app15147883

AMA Style

Kahveci S, Avaroğlu E. An Adaptive Underwater Image Enhancement Framework Combining Structural Detail Enhancement and Unsupervised Deep Fusion. Applied Sciences. 2025; 15(14):7883. https://doi.org/10.3390/app15147883

Chicago/Turabian Style

Kahveci, Semih, and Erdinç Avaroğlu. 2025. "An Adaptive Underwater Image Enhancement Framework Combining Structural Detail Enhancement and Unsupervised Deep Fusion" Applied Sciences 15, no. 14: 7883. https://doi.org/10.3390/app15147883

APA Style

Kahveci, S., & Avaroğlu, E. (2025). An Adaptive Underwater Image Enhancement Framework Combining Structural Detail Enhancement and Unsupervised Deep Fusion. Applied Sciences, 15(14), 7883. https://doi.org/10.3390/app15147883

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop