面向质量提升的水下图像增强集成融合模型

Ensemble Fusion of Underwater Image Enhancement(EFUIE)Model for Comprehensive Quality Improvement

水下图像存在因介质散射和吸收而引起的颜色失真、能见度低等问题,极大地限制了水下图像的应用。传统的水下图像增强模型很难同时实现色彩修正、对比度提升和去模糊,导致增强后的图像适应性不强。为此,提出了一种面向综合质量提升的水下图像增强集成融合模型(EFUIE),以全面提升水下图像质量。将原始图像分别通过MSRCR,Sea Thru,CLAHE和UDCP 4种图像增强算法预处理,然后利用主成分分析作为融合规则将小波高低频信息进行融合,从而将4种算法的权重进行有效集成,最终得到优化后的增强图像。在UIEB数据集上与目前较流行的11种水下图像增强模型结果进行比较,Entropy值、UIQM值和平均梯度值相对于性能第二的模型分别提高了0.844,1%和2.381。所提出的EFUIE模型,集成了传统水下图像增强算法的优势,在图像色彩、对比度、清晰度方面均有提升。

Underwater images usually have color distortion and low visibility problems caused by medium scattering and absorption,greatly limiting its application.Meanwhile,the traditional underwater image enhancement models are challenging to simultaneously achieve the color correction,contrast improvement and blur removal,resulting in the weak adaptability of the enhanced images.Therefore,an ensemble fusion of underwater image enhancement(EFUIE)model was proposed,aiming at comprehensively improving the image quality.In the proposed EFUIE,the original underwater images were separately enhanced by MSRCR,Sea Thru,CLAHE,and UDCP,respectively.Further,the temporally generated images were separately transformed by wavelet transform(WT),generating the corresponding approximate and detailed wavelet coefficients.Finally,the wavelet coefficients at high and low frequencies from the temporally generated images generated by the aforementioned four algorithms were fused by principal component analysis(PCA),to obtain the final the optimized images.The experimental test was conducted on the public UIEB dataset to compare with 11 currently popular image enhancement methods,Entropy values,UIQM values and mean gradient values improved by 0.844,1%and 2.381 relative to performance 2,respectively.The EFUIE proposed in this paper has simultaneously integrated the advantages of traditional underwater image enhancement models,and provides improvements in image color,contrast,and sharpness,comprehensively.