基于块效应抑制的压缩降质模糊图像盲复原

针对被压缩重构后的模糊图像复原问题，提出基于块效应抑制的图像去模糊方法.采用多尺度分解逐层进行模糊核估计，每一层利用去块效应的模糊图像作为有效显著边缘估计的参考图像，并在模糊核估计优化函数中加入模糊核梯度约束保证其连续平滑性，通过在频域交替迭代优化确定模糊核.最终复原约束项为图像全变分以及L2范数块效应抑制项，该项通过迭代使得复原图像和去块效应图像逐步逼近.实验结果表明，提出的方法能够在较高压缩率的模糊图像中估计出相对准确的模糊核，得到兼顾消除块效应与保留图像细节的复原效果.
Abstract: An image deblurring algorithm based on block effect suppression was proposed for compressed degraded blur image. Multi-scale decomposition was adopted layer by layer in the kernel estimation stage. The blurred image whose block artifacts were removed was used as the reference image in the effective significant edge estimation in each layer. Then the kernel gradient norm constraint was added in the kernel estimation optimization function to ensure that the kernel's shape was continuous and smooth. The kernel was determined by alternately iterative optimization in frequency domain. Final restoration constraints were image total variation and the L2 norm block suppression terms. The L2 term made the restoration image and the block effect eliminated image approach during iteration. Experimental results show that the proposed method can estimate relatively accurate kernel even when the blur image compression rate is high, and can acquire a satisfied restored image with block effect eliminated and image details preserved.