中文说明：我们提出了一种从单一图片去除运动模糊的算法。我们的方法在去模糊图像的计算过程中，对于卷积核的估计和清晰图像，采用统一的概率模型。我们分析了当前去模糊方法中通常存在的人工痕迹的产生原因，而后在我们的概率模型中引入了一些新的术语。这些术语包括模糊图像噪声的空域随机模型，还有新的局部平滑先验知识。通过对比度约束，即使是低对比的模糊图像，也能减少人工振铃效应。最后，我们描述了一种有效的优化方案，通过交替估计模糊核和清晰图像的复原过程直到收敛。经过这些步骤，我们能够在一个低的计算复杂度的时间内获得一个高质量的清晰图像。我们的方法生成的图像质量相当于用多张模糊图片生成的清晰图片的效果，而后者的方法需要额外的硬件资源。
English Description：
We propose an algorithm to remove motion blur from a single image. In the process of deblurring image calculation, our method uses a unified probability model for convolution kernel estimation and clear image. In this paper, we analyze the causes of artifical traces in current deblurring methods, and then introduce some new terms into our probability model. These terms include spatial stochastic model of blurred image noise and new local smoothing prior knowledge. By contrast constraint, the artificial ringing effect can be reduced even for low contrast blurred images. Finally, we describe an effective optimization scheme by alternately estimating the blur kernel and the restoration process of the clear image until convergence. After these steps, we can get a high quality clear image in a low computational complexity time. The image quality generated by our method is equivalent to the effect of clear image generated by multiple blurred images, while the latter method requires additional hardware resources.