发布时间: 2018-11-16 摘要点击次数: 全文下载次数: DOI: 10.11834/jig.180255 2018 | Volume 23 | Number 11 NCIG 2018会议专栏

 收稿日期: 2018-04-12; 修回日期: 2018-06-12 基金项目: 上海市自然科学基金项目（18ZR1400300） 第一作者简介: 孙晓帆, 1993年生, 女, 硕士研究生, 主要研究方向为智能图像处理与分析、图像质量评价等。E-mail:602269155@qq.com;张鑫生, 男, 硕士研究生, 主要研究方向为压缩域图像分析。E-mail:859095270@qq.com;吴乐明, 男, 硕士研究生, 主要研究方向为图像集压缩及评价。E-mail:569911638@qq.com;况奇刚, 男, 硕士研究生, 主要研究方向为图像超分辨率。E-mail:1148002867@qq.com. 中图法分类号: TP391.41 文献标识码: A 文章编号: 1006-8961(2018)11-1759-09

# 关键词

Consistent enhancement assessment for an underwater image set
Sun Xiaofan, Liu Hao, Zhang Xinsheng, Wu Leming, Kuang Qigang
College of Information Science and Technology, Donghua University, Shanghai 201620, China
Supported by: Shanghai Municipal Natural Science Foundation (18ZR1400300)

# Abstract

Objective An increasing number of underwater image enhancement methods have been put into practical applications because underwater images typically have quality degradation problems, such as blurring, distortion, and low visibility. At present, each quality evaluation criterion mainly focuses on the single image. Existing methods adopt the average quality score of a quality evaluation criterion as an indicator, and the enhancement algorithm is evaluated by the average score. However, the non-consistent average quality score changes with the image set and produces large fluctuations. If an enhancement algorithm cannot consistently improve the image quality score in a small-scale image set, then the average quality score has certain limitations and large error when the enhancement algorithm is applied to a large-scale image set. To solve the abovementioned problems, a universal underwater image quality assessment method, namely, consistent enhancement quality assessment (CEQA) for an underwater image set, is proposed. Method The proposed method can judge the consistency of the enhancement algorithm by comparing the difference of the quality score before and after image enhancement and by changing the weight proportion of the selected quality score difference, unifying the fractional system, and calculating the CEQA fraction of the enhanced image set. The concrete steps of this proposed method are as follows:1) An image set ({ ${{I}_{1}},\; {{I}_{2}}, \;{{I}_{3}}, \;\cdots , \;{{I}_{n}}$}, where $n$ is the total number of images of the underwater image set) is determined, and then a quality evaluation criterion M is selected to evaluate the image quality of the original underwater image $I_1$ to obtain a quality score ${{\alpha }_{1}}$ of the original image $I_1$. 2) The proposed method can process the original underwater image $I_1$ through the image quality enhancement algorithm A, and obtain the enhanced image ${{{{I}'}}_{1}}$. 3) the proposed method uses the quality evaluation criterion M, which is used in Step 1, to evaluate the quality of the enhanced image ${{{{I}'}}_{1}}$ and obtain the quality score ${{\beta }_{1}}$. 4) the quality score ${{\beta }_{1}}$ is subtracted by the quality score ${{\alpha }_{1}}$ to obtain the fractional difference $Q_1$. 5) Steps 1-4 are successively performed for the original underwater images $I_2$, $I_3$, …, $I_n$ to obtain fractional difference ${{Q}_{2}},\; {{Q}_{3}},\; \cdots ,\; {{Q}_{n}}$, correspondingly. If the ${{Q}_{1}},\; {{Q}_{2}},\; \cdots , \;{{Q}_{n}}$ values are all positive, then the underwater image quality enhancement algorithm A, under the quality evaluation criterion M, can consistently enhance the quality of this underwater image set, and then Step 6 is performed. Otherwise, the quality enhancement algorithm A is an inconsistent quality enhancement algorithm under these conditions. 6) the maximum value of ${{Q}_{1}}, \;{{Q}_{2}}, \;\cdots , \;{{Q}_{n}}$ is selected as $Q_{\rm max}$ and the minimum value as $Q_{\rm min}$. The average value of ${{Q}_{1}},\; {{Q}_{2}}, \;\cdots , \;{{Q}_{n}}$ is determined as $Q_{\rm ave}$. 7) the effective value $C_{\rm eff}$ of the underwater image quality enhancement algorithm A for this image set is obtained under the quality evaluation criterion M by normalizing the average value $Q_{\rm ave}$ and the minimum value $Q_{\rm min}$ and then adjusting its proportion. Under the selected quality evaluation criterion M, the non-consistent quality enhancement algorithm for the same underwater image set cannot consistently enhance the image set under the quality evaluation criterion M to evaluate different underwater image quality enhancement algorithms ${{{\rm A}}_{1}},\; {{{\rm A}}_{2}},\; \cdots ,\; {{{\rm A}}_{m}}$ ($m$ is the total number of the quality enhancement algorithms). By contrast, the consistent quality enhancement algorithm can effectively enhance the quality of the image set. If the average value $Q_{\rm ave}$ is different, then the quality enhancement algorithm with high effective value $C_{\rm eff}$ has significant enhancement strength and improved enhancement capability when comparing several consistent quality enhancement algorithms; if the average value $Q_{\rm ave}$ is the same, then the quality enhancement algorithm with high effective value $C_{\rm eff}$ has an improved stability. Result The experimental results of the quantitative analysis of the mean value method show that the average value is larger in UCIQE and entropy than in the original image after the image set is enhanced by the three image quality enhancement algorithms, which are randomly selected. However, the quality score is lower in numerous single images than in the original image. In the extended application of CEQA method, using the UCIQE evaluation criteria of a selected underwater image set, the enhancement effect of the CLAHE-HSV algorithm is optimal, and the inverse filtering algorithm is better than the filtering-guided dark channel defogging algorithm. Many experimental data show that our method can effectively solve these problems and provide an evaluation criterion for the quality enhancement algorithm of the image set. The comparative experimental results between the CEQA and the mean value methods show that the non-consistent quality enhancement algorithm has the highest mean value when the image set is small, but its mean value is lower than that of the original image when the image set is enlarged. Therefore, the inconsistent quality enhancement algorithm has an extensive or serious reduction of image quality. The consistent quality enhancement algorithm before and after expanding the image set can steadily improve the image quality, and thus the average value of the quality score is consistently higher than the mean value of the original image. Conclusion The experimental results of the extended application of the CEQA method show that the proposed method is feasible and can obtain effective experimental data to compare the advantages and disadvantages of underwater image quality enhancement algorithms. The experimental results of the comparison between the CEQA and average value methods suggest that the proposed method is more accurate than the average value method and effectively controls the large sample deviation. Therefore, a consistent enhancement assessment method for the underwater image quality is proposed; this method provides an improved evaluation criterion for underwater image quality enhancement algorithm in large-scale practical applications. The proposed consistent enhancement evaluation method is better than the existing mean value method for evaluating an image set and provides a quantifiable performance index for the new image quality enhancement algorithm. The proposed method has a guiding effect on the advantages and disadvantages of a new image quality enhancement algorithm in the future. In addition, the formula of this method is simple, universal, highly flexible, and easy to understand; this formula can be applied to various fields of image quality evaluation. The shortcoming of the proposed method includes a high requirement for robustness and stability of an enhancement algorithm. The formula is suitable for the applications in the zero-fault-tolerance field. A reliable quality enhancement algorithm is selected for applications with stringent performance requirements. For common application requirements, the standard performance of this method is relatively high, and several algorithms without fluctuation cannot satisfy the requirements of this consistent enhancement assessment method. The underwater image enhancement technology can still be developed, and the enhancement performance requires added authoritative evaluation criterion. Future research should focus on developing an additional fault-tolerant method, and a favorable quality enhancement algorithm can be selected for different application requirements when facing a certain application. An application can select a quality enhancement algorithm under strict conditions, can change the screening conditions in accordance with the specified requirements, and obtain the specific experimental data and results.

# Key words

image set; consistent enhancement; image quality; image enhancement; quality assessment

# 1.1 单幅图像的质量评价准则

Yang和Arcot提出了质量评价准则UCIQE(underwater colour image quality evaluation)[1]，该准则是目前应用最广泛的水下图像质量评价准则，UCIQE通过线性组合饱和度、色度与对比度来对水下图像的模糊、非均匀色差和对比度进行量化评估。香农(Shannon)[2]提出信息熵将热力学概率扩展到系统中各个信息源信号出现的概率。通常情况下，信息熵越大，说明该质量增强算法更加有效地减少了图像信息丢失，并更好地增加了有价值的信息[3]

# 2.2 平均值方法的定量分析

Table 1 The comparison of average value of UCIQE and entropy

 增强前后 原图 直方图均衡 CLAHE 无监督颜色模型 UCIQE 0.52 0.62 0.61 0.59 熵 7.35 7.43 7.62 7.49

# 3 本文方法

CEQA方法的具体步骤为：

1) 首先确定某一图像集{${{\mathit{\boldsymbol{I}}}_{1}}, {{\mathit{\boldsymbol{I}}}_{2}}, {{\mathit{\boldsymbol{I}}}_{3}}, \cdots , {{\mathit{\boldsymbol{I}}}_{n}}$}($n$为该水下图像集的图像总数量), 再选取一种质量评价准则$M$对水下获取的原始图像${{\mathit{\boldsymbol{I}}}_{1}}$的图像质量进行评价，得出一个原图像${{\mathit{\boldsymbol{I}}}_{1}}$的质量分数${{\alpha }_{1}}$，作为公式中的参数之一，也作为评价图像质量增强算法好坏的基准。其中，图像集{${{\mathit{\boldsymbol{I}}}_{1}}, {{\mathit{\boldsymbol{I}}}_{2}}, {{\mathit{\boldsymbol{I}}}_{3}}, \cdots , {{\mathit{\boldsymbol{I}}}_{n}}$}与质量评价准则$M$为实验变量。

2) 接下来，将这幅水下获取的原始图像${{\mathit{\boldsymbol{I}}}_{1}}$通过图像质量增强算法$A$处理，得到增强后的图像${{{\mathit{\boldsymbol{{I}'}}}}_{1}}$，其中，质量增强算法$A$为实验对象，是该次验证评价的主要算法。

3) 使用步骤1)中使用的质量评价准则$M$对增强后的图像${{{\mathit{\boldsymbol{{I}'}}}}_{1}}$做质量评价，得到质量分数${{\beta }_{1}}$，若增强后的图像质量分数${{\beta }_{1}}$高于基准值${{\alpha }_{1}}$，则${{Q}_{1}}$值为正，可以说明在质量评价准则$M$的检测下，图像质量增强算法$A$可以对这一幅原始图像${{\mathit{\boldsymbol{I}}}_{1}}$增强。

4) 将得到的质量分数${{\alpha }_{1}}$${{\beta }_{1}}$代入求${{Q}_{i}}$

 ${{Q}_{i}}={{\beta }_{\mathit{i}}}-{{\alpha }_{\mathit{i}}}$ (1)

5) 依次对水下获取的原始图像${{\mathit{\boldsymbol{I}}}_{2}}, {{\mathit{\boldsymbol{I}}}_{3}}, \cdots , {{\mathit{\boldsymbol{I}}}_{n}}$重复步骤1)—4)，分别得到${{Q}_{2}}, {{Q}_{3}}, \cdots , {{Q}_{n}}$分数。若${{Q}_{i}}$($i$=1，2，…，$n$)值全为正，则说明，这一水下图像质量增强算法$A$在质量评价准则$M$的评价条件下，能够对该水下图像集进行一致性增强，则继续进行步骤6)，否则，得出结论：该质量增强算法$A$为非一致性增强的。

6) 首先求出${{Q}_{i}}$的最大值

 ${{Q}_{{\rm max}}}={\rm max}\left\{ {{Q}_{1}}, \cdots , {{Q}_{\mathit{i}}}, \cdots , {{Q}_{n}} \right\}$ (2)

 ${{Q}_{{\rm min}}}={\rm min}\left\{ {{Q}_{1}}, \cdots , {{Q}_{\mathit{i}}}, \cdots , {{Q}_{n}} \right\}$ (3)

 ${{Q}_{{\rm ave}}}={\rm ave}\left\{ {{Q}_{1}}, \cdots , {{Q}_{\mathit{i}}}, \cdots , {{Q}_{n}} \right\}$ (4)

7) 得到在质量评价准则$M$下，这一水下图像质量增强算法$A$对该图像集的${C_{{\rm{eff}}}}$值，定义为

 $\begin{array}{l} {C_{{\rm{eff}}}} = \lambda \frac{{{Q_{{\rm{ave}}}}}}{{{Q_{{\rm{max}}}} - {Q_{{\rm{min}}}}}} + \\ \;\;\left( {1{\rm{ - }}\lambda } \right)\frac{{{Q_{{\rm{min}}}}}}{{{Q_{{\rm{max}}}} - {Q_{{\rm{min}}}}}} \end{array}$ (5)

# 4.1 CEQA方法的扩展应用

Table 2 The consistent enhancement ${C_{{\rm{eff}}}}$ value

 图像质量增强算法 逆滤波 导向滤波暗通道去雾 CLAHE-HSV ${C_{{\rm{eff}}}}$ 1.66 1.62 2.82

# 4.2 CEQA方法与平均值方法的对比实验

Table 3 The ${C_{{\rm{eff}}}}$ value compared with average value (by using the UCIQE criterion)

 方法 图像集 原图 动态阈值白平衡 对比度拉伸 平均值 图像集$\mathit{\boldsymbol{a}}$ 0.420 0.426 0.425 图像集$\mathit{\boldsymbol{b}}$ 0.427 0.407 0.434 ${C_{{\rm{eff}}}}$ 图像集$\mathit{\boldsymbol{a}}$ 无 非一致性增强 19.168 图像集$\mathit{\boldsymbol{b}}$ 7.652

# 参考文献

• [1] Yang M, Sowmya A. An underwater color image quality evaluation metric[J]. IEEE Transactions on Image Processing, 2015, 24(12): 6062–6071. [DOI:10.1109/TIP.2015.2491020]
• [2] Guo J C, Li C Y, Zhang Y, et al. Quality assessment method for underwater images[J]. Journal of Image and Graphics, 2017, 22(1): 1–8. [郭继昌, 李重仪, 张艳, 等. 面向水下图像的质量评价方法[J]. 中国图象图形学报, 2017, 22(1): 1–8. ] [DOI:10.11834/jig.20170101]
• [3] Li C Y, Guo J C, Chen S J, et al. Underwater image restoration based on minimum information loss principle and optical properties of underwater imaging[C]//Proceedings of 2016 IEEE International Conference on Image Processing. Phoenix, AZ, USA: IEEE, 2016.[DOI:10.1109/ICIP.2016.7532707]
• [4] Wu X Y, Wu J, Zhang H. Research on image restoration techniques based on inverse filtering algorithm[J]. Information Technology, 2011, 35(10): 183–185. [吴雪垠, 吴谨, 张鹤. 逆滤波法在图像复原中的应用[J]. 信息技术, 2011, 35(10): 183–185. ] [DOI:10.3969/j.issn.1009-2552.2011.10.050]
• [5] He S, Pan X L, Li Y M. Optimization algorithm for average filtering[J]. Information Technology, 2012, 36(3): 133–134, 137. [何石, 潘晓璐, 李一民. 一种均值滤波的优化算法[J]. 信息技术, 2012, 36(3): 133–134, 137. ] [DOI:10.13274/j.cnki.hdzj.2012.03.041]
• [6] Li Y J, Su H Q, Yang F, et al. Improved algorithm study about removing image noise[J]. Computer Engineering and Design, 2009, 30(12): 2995–2997. [李彦军, 苏红旗, 杨峰, 等. 改进的中值滤波图像去噪方法研究[J]. 计算机工程与设计, 2009, 30(12): 2995–2997. ] [DOI:10.16208/j.issn1000-7024.2009.12.030]
• [7] [陈尔言, 王卫静. 基于导向滤波的暗通道去雾算法研究[J]. 科技展望, 2016, 26(4): 174–175. ] [DOI:10.3969/j.issn.1672-8289.2016.04.163]
• [8] Tarel J P, Hautière N. Fast visibility restoration from a single color or gray level image[C]//Proceedings of the 12th International Conference on Computer Vision. Kyoto, Japan: IEEE, 2009: 2201-2208.[DOI:10.1109/ICCV.2009.5459251]
• [9] Su Z F. Studying and implementation of image signal preprocessing based on FPGA[D]. Guangzhou: South China University of Technology, 2015. [苏志锋.基于FPGA的图像预处理研究与实现[D].广州: 华南理工大学, 2015.] http://cdmd.cnki.com.cn/Article/CDMD-10561-1015987964.htm
• [10] Li Y H, Liu B J. The image enhancement based on histogram equalization[J]. Journal of North China Institute of Science and Technology, 2003, 5(2): 65–67. [李耀辉, 刘保军. 基于直方图均衡的图像增强[J]. 华北科技学院学报, 2003, 5(2): 65–67. ]
• [11] Hitam M S, Awalludin E A, Yussof W N J H J, et al. Mixture contrast limited adaptive histogram equalization for underwater image enhancement[C]//Proceedings of 2013 International Conference on Computer Applications Technology. Sousse, Tunisia: IEEE, 2013: 1-5.[DOI:10.1109/ICCAT.2013.6522017]
• [12] Chen Y, Huo F R, Miao H. Application of contrast stretching in optical correlation detection and recognition[J]. Chinese Journal of Scientific Instrument, 2008, 29(4): 795–798. [陈宇, 霍富荣, 苗华. 对比度拉伸在目标探测与识别中的应用研究[J]. 仪器仪表学报, 2008, 29(4): 795–798. ]
• [13] Yang Y, Guo L, Wang T J. Multi-scale structure tensor based unsupervised color-texture image segmentation approach in multiclass[J]. Journal of Computer-Aided Design & Computer Graphics, 2014, 26(5): 812–825. [杨勇, 郭玲, 王天江. 基于多尺度结构张量的多类无监督彩色纹理图像分割方法[J]. 计算机辅助设计与图形学学报, 2014, 26(5): 812–825. ]
• [14] Jian M W, Lam K M, Dong J Y, et al. Visual-patch-attention-aware saliency detection[J]. IEEE Transactions on Cybernetics, 2015, 45(8): 1575–1586. [DOI:10.1109/TCYB.2014.2356200]
• [15] Jian M W, Qi Q, Dong J Y, et al. The OUC-Vision large-scale underwater image database[C]//Proceedings of 2017 IEEE International Conference on Multimedia and Expo. Hong Kong, China: IEEE, 2017.[DOI:10.1109/ICME.2017.8019324]
• [16] Jian M W, Qi Q, Dong J Y, et al. Saliency detection using quaternionic distance based weber local descriptor and level priors[J]. Multimedia Tools and Applications, 2018, 77(11): 14343–14360. [DOI:10.1007/s11042-017-5032-z]