基于梯度先验的水下图像恢复

Underwater image restoration based on gradient prior

由于水体对光线的吸收和散射作用,水下图像呈现出强衰减、高噪声和色彩畸变等问题,难以满足视觉观察和图像分析的需求。针对这一问题,提出了一种基于梯度先验的水下图像恢复方法,用以提高水下图像质量。首先,根据水下成像特性,建立水下图像梯度先验,水下图像中目标反射信息(水下清晰图层)的梯度大于散射噪声信息(水下噪声图层);其次,根据水下成像模型,对上述梯度先验进行表征建模;最后,建立水下图像的梯度分布优化函数,采用半二次优化方法分离出目标反信息,作为水下图像恢复结果。以UEIB和RUIE数据集为实验样本,与近年来所提出的5种水下图像处理方法进行比较,方法在定性和定量两类评价中均获得了出色的处理结果,峰值信噪比(PSNR)、结构相似性(SSIM)和水下图像质量评价指标(UIQM)评价指标分别达到20.0665、0.6961和3.9029,均优于对比方法。因此,该方法能够有效地抑制水下图像噪声,提高水下图像的信噪比、清晰度和对比度。

Due to the light absorption and scattering effect of the water volume,underwater images suffer from strong attenuation,high noise and serious color distortion.Thus,the original underwater images can hardly meet the requirements of visual observation and image analysis.To solve this problem,this article proposes an underwater image restoration method based on gradient prior for improving underwater image quality.First,according to the underwater imaging prosperity a gradient prior is established for underwater images that the gradient of scene imaging information(the clear layer of underwater images)must be larger than that of scattering information(the haze layer of underwater images).Secondly,the gradient prior is modeled with the underwater imaging model.Finally,a semi-quadratic optimization function is utilized to extract the object reflection information which generates final image restoration results.Underwater samples in UEIB and RUIE datasets are taken for experiment evaluation.In contrast to five state-of-the-art underwater image information processing methods,the proposed method obtained prominent performance in both qualitative and quantitative evaluation.The average PSNR,SSIM,and UIQM values of this method are 20.0665,0.6961,and 3.9029,which are better than compared methods.Therefore,the proposed method can efficiently remove underwater image noise.The image signal to noise ratio,sharpness and the contrast of underwater images are enhanced.