结合颜色补偿和卷积神经网络去雾模型的水下图像色彩恢复研究

Color restoration of underwater images using color compensation and convolutional neural network based defogging model

水下视觉测量技术已被广泛应用于海洋测绘、水下精密工程测量、水下考古和水下环境监测等领域。然而水下图像存在颜色失真、图像模糊和对比度低的问题,限制了水下视觉测量技术在实际环境中的应用。针对这些问题,本文提出了一种结合颜色补偿和卷积神经网络(CNN)去雾模型的水下图像色彩恢复方法,分步对水下图像进行增强处理。首先分析水下图像色偏情况,采用自适应的颜色补偿策略,结合灰度世界白平衡算法,实现水下图像的色彩校正;然后根据水下图像特点,设计了基于CNN的去雾模型,通过对综合性参数的估计,实现对水下图像的去雾处理;最后采用限制对比度的自适应直方图均衡化CLAHE方法增强水下图像对比度。结果表明,与其他经典水下图像增强算法相比,在处理各种环境下的水下图像时,该方法成功地提高了图像的清晰度,降低了受损水下图像的色彩偏差,具有优越的水下图像色彩恢复能力。

Underwater vision measurement has important applications in marine surveying,underwater engineering surveying,underwater archaeology and underwater environmental monitoring.However,underwater images suffer from color distortion,image blurring and low contrast,which limits the application of underwater visual measurement technology in practical environments.A color restoration method for underwater images based on color compensation and convolutional neural network(CNN)defogging model is proposed in this paper,in which the image enhancement is carried out step-by-step.Firstly,the color deviation of underwater images is analyzed,and then an adaptive color compensation strategy combined with the grayscale world white balance algorithm is used to correct underwater image color.Secondly,a CNN based dehazing model was designed to achieve dehazing processing of underwater images.Finally,the adaptive histogram equalization CLAHE method is used to enhance the contrast of underwater images.In order to prove the applicability and superiority of the proposed method,two image datasets are combined to study,and several known underwater image enhancement and restoration methods are compared.The proposed method and several compared methods are evaluated in two aspects of subjective visual effect and quantitative evaluation index.The comparison results show that compared with other enhancement algorithms,the proposed method successfully improves the clarity of the image and reduces the color deviation of the damaged underwater image when processing underwater images in various environments and has superior image color recovery compared with existing enhancement methods.