基于改进湍流模型和偏振成像技术的水下退化图像复原方法

Restoring method for underwater degraded images based on improved turbulence model and polarization imaging

为了满足工厂化水产养殖中过程控制的信息化要求,该文提出一种基于改进湍流模型结合偏振成像技术,具有较强鲁棒性的水下退化图像复原方法。考虑内外尺度对波结构函数影响,结合折射率谱,改进水下湍流退化模型以提高复原算法先验知识的完备性;基于改进的退化模型和水下前向散射光的偏振特性,利用偏振成像技术提取退化图像中的噪声特征;基于退化图像噪声特征,采用约束最小二乘滤波法进行退化图像复原;最后,对复原效果进行相应比较,结果表明在强湍流条件下本算法具有更为理想的复原效果。该研究可为复杂水流条件下水下退化图像复原方法研究提供参考。

Eestimating their weight and size, therefore valuing their growing steps, based on snapped their photographs, is an efficient and quick method in order to meet the information requirment for process controlling in industrial aquaculture. For this purpose, effective restoring methods to process underwater degraded images are essential for this technology. Based on polarization imaging technology and improved turbulence model, a robust model for underwater degraded images restoration was proposed in this paper. Firstly, aimed at increasing the completeness of prior knowledge required by restoring degraded images method, an improved underwater turbulence model was designed by considering the wave structure function and distribution function of scattering scale parameter to overcome the shortcoming resulting from the simplified turbulence model, which simply imitated the situation for atmospheric turbulence and graded turbulence intensity only depending on fuzzy factor. Secondly, noise characteristics in degraded images was drawn from comparison between ＂and ＂ , ＂minus＂ images, which were computed from polarization images, based on polarization characteristics of underwater forward scattering light in visible band. For the precision of noise characteristics drawing, pulses coupled neural network （PCNN） and wavelet transfer （WT） algorithm was applied in computing the ＂and ＂ and ＂minus＂ images. And then, the algorithm, namely constrained least squares filtering （CLSF） method, was applied to replace Wiener filtering for restoring the degraded images because of its robust characteristics. Finally, comparison of restoration results among the four different restoring methods was carried out to evaluate the effect of our proposed method according to four valuating parameters, namely average value （AVG）, standard deviation （SD）, Entropy and signal-to-noise ratio（SNR）. The experimental conditions, especially for turbulence situations, were designed as： feeding Chinese carp with weight of about 0.5 kg at a pool （5 m×4 m×2 m）. All images were snapped at 30 cm depth under water surface. The results showed that more ideal restoration effect in strong turbulence circumstances could be realized based on the proposed method, i.e. improved turbulance model accompanied with CLSF This should be beneficial for further research works on underwater degraded images restoration in complex flow conditions.