中文摘要：

由于在相机拍摄的人脸中往往会存在较大的模糊,为了有效去除人脸图像的抖动模糊,提出一种基于非均匀去模糊与人脸特殊属性相结合的人脸图像去模糊方法。首先研究非均匀去模糊的原理,并提出通过约束相机运动子空间的方法来估计出相机与人脸之间的相对运动路径。再根据人脸的特殊属性,通过对清晰人脸训练得到一组清晰的人脸字典,建立人脸的先验知识。最后利用得到的非均匀模糊核和人脸字典对模糊人脸图像进行去卷积。实验结果表明,提出的方法相对于现有的去模糊算法可以得到更清晰的人脸图像,对后续的人脸识别有很大的辅助作用。

英文摘要：

Since the faces to be shot by camera will usually have bigger blur,we proposed a face image deblurring method,which is based on combining the nonuniform deblurring with special face property,for effectively removing the dithering blurs on face image. First,we studied the principle of nonuniform deblurring,and proposed to estimate the relative motion path between camera and face by the method of constraining the subspace of camera movement. Then,according to the special property of face,we obtained a group of clear face dictionary through training clear faces and established the priori knowledge of faces. Finally,we employed the derived nonuniform blur kernel and face dictionary for convolution on blurring face image. Experimental results indicated that the proposed method was able to get clearer face image relative to existing deblurring algorithm. This plays a great auxiliary role on subsequent face recognition.