在使用机器视觉对家电产品的标签进行品质检测的过程中，标签图像的清晰度 是保证视觉检测顺利进行的关键。为了解决由固定在机器人末端的运动相机拍摄的标签 图像存在运动模糊的问题，提出一个基于 L0 范数的正则化模型。对于模糊的图像，一 般的去模糊算法已经探索并使用了大量的图像先验，可以恢复良好的整体视觉效果，但 一般的去模糊算法没有充分考虑标签的特征，而这是重要的去模糊先验。针对这一问 题，对标签图像的像素分布和梯度分布特征进行分析，提出了基于稀疏梯度先验的正则 化模型。实验结果表明，与其他去模糊算法相比，对于合成图像和真实图像，文中提出 的方法在恢复图像清晰度的同时，可有效地抑制图像边缘的振铃效应，运算速度较之前 提高了 80. 52% 。

Label image sharpness is the key point that influences the effectiveness of visual inspection during the label qualify inspection of household appliances with machine vision． To solve the motion blurring problem of tag image captured by camera fixed on the moving robot，a regularization model based on L0 norm was proposed． The general methods of deblurring algorithms have considered and implemented much of image priors for blurred images，which can obtain the well-done visual overall effect． However，these general algorithms does not take the characteristic of identification tag into consideration，and this is an important prior to deblurring． To this point，a specific model consisting of image gradient prior regularization and sparse regularization was proposed to analyze pixel distribution and gradient distribution features of label image． The experimental results show that，as compared with other deblurring algorithms，the proposed method can effectively suppress the ringing effect at the edge of the label image while restoring label image sharpness on synthetic images and real images，and the real computation speed is increased by 80. 52% ．