运动模糊的图像复原,是图像处理和计算机视觉领域中的热点问题。通过分析图像的纹理特征,提出一种鲁棒的显著性边缘提取方法,并将其应用于运动模糊图像恢复中。联合图像纹理的梯度幅值、方向信息,优化模糊图像并抑制噪声干扰,从而提取稳定的强显著性边缘。在模糊核估计的能量模型中,引入梯度约束方法,保护模糊核的结构特征。对于较大尺寸的模糊核,采用多尺度策略进行图像复原。实验结果表明,文中算法能够有效提高模糊核估计的准确性,提升运动模糊图像复原的质量。
Motion deblurring is hotly discussed in the computer vision and graphics community. The classical hard thresholding method is found destroying the image edges. In this paper, we put forward a robust motion deblurting method based on the analysis of image textures. First, we developed a novel method for computing image structures based on gradient magnitude and orientation so that the local structure could be protected to the most extent. Second, we constructed a gradient constraint by mitigating the possible adverse effect of salient edges to improve the robustness of kernel estimation. Third, the multiresolution technology was used when the blur kernel was large. Preliminary numerical results show that the proposed method can effectively remove the motion blur and have a better effect on the image edge preservation.