针对运动过程中视觉图像易产生运动模糊的问题, 提出了一种基于稀疏表示和Weber定律相结合的图像盲复原方法。该方法利用冲击滤波器预测模糊图像的显著边缘梯度, 并用多尺度策略由粗到细进行模糊核的估计。然后, 对图像盲复原模型进行稀疏正则化约束, 并结合反映人类视觉特性的Weber定律对合成模糊图像和真实模糊图像进行盲复原。实验结果表明, 本文采用的盲复原算法的性能指标和图像的纹理都达到了较优的复原效果。与近年较好的Rob Fergus去模糊方法和Xu Li去模糊方法相比, 对Lena模糊图去模糊后的结构相似度(SSIM)为0.762 4, 峰值信噪比(PSNR)提高了1.82~2.99 dB; 对Cameraman模糊图去模糊后的结构相似度(SSIM)为0.8589, PSNR提高了2.46~5.58 dB。 另外, 本文方法降低了复原图像的边界伪影, 符合人的视觉感知特性。

For the motion blur problem of a visual image produced in moving processing, a blind image restoration method based on sparse representation and Weber's law is proposed. The method uses a shock filter to predict the sharp edges of blurred images, and a multi-scale strategy to estimate the blur kernel from a coarse estimation to a fine one. The sparse representation is treated as a priori knowledge for regularization constraint of blind image restoration model, and the Weber's law which reflects the human visual characteristics is combined to conduct blind restoration for the synthetic blurred image and the real blurred image. Experimental results show that the proposed method achieves better restoration results both for the performance indexes and the image textures. As compared with the Rob Fergus's method and Xu Li's method developed in recent years, it shows that the structural similarity (SSIM) is 0.762 4 and the Peak Signal to Noise Ratio (PSNR) is improved by 1.82 dB to 2.99 dB for the deblurred Lena image, and the SSIM is 0.858 9; and the PSNR has improved by 2.46 dB to 5.58 dB for the deblurred Cameraman image. Moreover, the proposed method reduces the boundary artifacts of the restored image, which is better consistent with human visual perception characteristics.