改进混合正则化约束多帧湍流退化图像盲复原方法

针对高速湍流造成成像平台接收目标图像模糊的问题,基于L₀正则化图像盲复原方法,提出了一种改进的混合正则化约束多帧湍流退化图像盲复原方法。首先,根据湍流退化时空变化关系,构建多帧退化图像复原模型描述湍流退化过程。其次,图像正则项在图像梯度L₀范数正则化基础上,增加图像梯度的L₂范数约束,改善复原图像中的阶梯伪像。再次,针对模糊核正则项,依据对湍流退化图像点扩散函数特性分析,提出了L₀-L₂混合正则化约束,保证了支持域的连续平滑特性。最后,使用多尺度图像金字塔的策略优化了求解过程。实验结果表明,该方法较好地复原湍流退化图像,与近年提出的具有代表性算法相比,在视觉效果和客观质量评价指标均有提升。

Improved mixed regularization constrained multi frame turbulence degradation image blind restoration

An improved mixed regularization constrained multi-frame turbulence degradation image restoration method is proposed to solve the problem that the target image is degraded by the turbulence imaging platform. Firstly, according to the turbulence degeneration spatiotemporal transformation relation, the multi-frame degeneration image restoration model is adopted. Secondly, the image regular term increases the L₂ gradient constraint of the image gradient on the basis of the normalization of the image gradient L₀norm, and improves the step artifacts in the restored image. Thirdly, according to the sparseness and smoothness of the image, the mixed regularization constraint is proposed. Finally, the multi-scale image pyramid strategy is used to improve the image restoration algorithm and optimize the solution process. The experimental results show that the method is effective in recovering the turbulence degradation image, and it has improved the visual effect and the objective quality evaluation index compared with the representative algorithm proposed in recent years.