融合深度学习与多尺度Retinex的水下图像增强方法

Combining deep learning and multi-scale Retinex for underwater images enhancement

针对水下图像纹理模糊和色偏严重等问题,提出了一种融合深度学习与多尺度导向滤波Retinex的水下图像增强方法。首先,将陆上图像采用纹理和直方图匹配法进行退化,构建退化水下图像失真的数据集并训练端到端卷积神经网络(convolutional neural network,CNN)模型,利用该模型对原始水下图像进行颜色校正,得到色彩复原后的水下图像;然后,对色彩复原图像的亮度通道,采用多尺度Retinex(multi-scale Retinex,MSR)方法得到纹理增强图像;最后,融合色彩复原图像中的颜色分量和纹理增强图像得到最终水下增强图像。本文利用仿真水下图像数据集和真实水下图像对提出方法进行性能测试。实验结果表明,所提方法的均方根误差、峰值信噪比、CIEDE2000和水下图像质量评价指标分别为0.3020、17.2392 dB、16.8784和4.9600,优于5种对比方法,增强后的水下图像更加真实自然。本文方法在校正水下图像颜色失真的同时,能有效提升纹理清晰度和对比度。

An underwater image enhancement method combining deep learning and multi-scale orientation filter Retinex is proposed to tackle the problems of blurry texture and serious color distortion.Firstly,the land image is degraded by texture and histogram matching method to establish a dataset which simulates the underwater image distortion,and an end-to-end convolutional neural network(CNN)model is built.By using the model,color correction is performed on original underwater images to obtain color-restored underwater images.Then,the multi-scale Retinex(MSR)method is used for the brightness channel of the color restoration images to generate texture-enhanced images.Finally,chrominance of the color-restored images and the texture-enhanced images are fused to eventually get the enhanced underwater images.The proposed method is tested on the simulated underwater image dataset and real underwater images individually.The experimental results show that root mean square error,peak signal-to-noise ratio,CIEDE2000,and underwater image quality measurement are 0.3020,17.2392 dB,16.8784 and 4.9600 and prevail to five comparison methods.The enhanced underwater images are more real and natural.In conclusion,the proposed method can effectively improve the clarity and contrast while accurately correcting the color distortion of the underwater images.