双头增强与非均匀校正的水下图像增强算法

Dual-head enhancement and non-uniform correction for underwater image enhancement algorithm

受水下环境浑浊度高、光照不足和均匀性差等因素影响,水下成像所获取的图像存在对比度低、细节模糊和颜色失真等缺陷,为此提出一种双头增强和非均匀校正的水下图像增强算法。利用构建的双头增强网络从浅层信息中提取多尺度特征,同时融合不同通道的上下文信息,有利于水下图像低对比度的增强;此外构建了非均匀校正网络,对图像不同通道和不同位置进行非线性加权融合,有利于颜色一致性和亮度的恢复。与10种算法相比,本文算法在UIEB测试集上峰值信噪比、结构相似性比其他方法的最优值分别提高了4.02 dB和0.120,CIEDE2000指标下降了1.51,在LUSI测试集上述指标分别提高了2.13 dB、0.025及下降了0.48。实验结果表明:所提算法针对不均匀的水下图像增强效果显著,更加符合人眼特性。

Affected by the high turbidity,insufficient illumination and poor uniformity of the underwater environment,the images obtained by imaging mechanisms have defects such as low contrast,blurred details and color distortion.To handle above problems,an underwater image enhancement algorithm based on a dual-head enhancement and non-uniform correction was proposed.The dual-head enhanced network was constructed to extract multi-scale features from shallow information and to fuse the context information of different channels,which was conducive to the enhancement of low contrast of underwater images.Furthermore,the constructed non-uniform correction network was used to perform nonlinear weighted fusion of different channels and positions of the image,which was conducive to the recovery of color consistency and brightness.Compared with the 10 algorithms,the optimal value of the peak signal-to-noise ratio and structural similarity in the UIEB test set was improved by 4.02 dB and 0.120,decreased by 1.51 on the CIEDE2000 index,and decreased by 2.13 dB,0.025 and 0.48 on the LUSI test set,respectively.Experimental results show that the proposed algorithm has a significant enhancement effect for non-uniform underwater images,and is more in line with the characteristics of the human eye.