基于归一化超拉普拉斯先验项的运动模糊图像盲复原

Blind image restoration based on normalized hyper laplacian prior term

基于变分方法提出了一种运动模糊退化图像的盲复原算法。考虑自然场景的图像梯度符合长拖尾概率分布,提出的方法采用归一化的超拉普拉斯先验项作为变分能量方程中的光滑项,从而有利于图像在去模糊的求解过程中正确解收敛。由于建立的能量方程不是严格凸的函数,故引入了分裂方法进行求解。整个运动模糊退化图像的盲复原过程在多尺度框架下由粗到细尺度渐进执行。最后利用估计出的点扩展函数计算清晰图像。相对于传统的盲复原算法,本文提出的算法不需要预测图像的梯度信息和对梯度进行筛选,直接求解能量方程就能够得到相应的正确解。得到的结果验证了本文算法的有效性。

An blind image restoration method for degraded motion images is proposed based on the variational method.As image gradient is coincident with the probability distribution of heavy-tailed characteristics in a natural scene,the method uses normalized hyper Laplacian prior term as a smooth term of variational energy equation to converge the solution of the equation exactly in image debluring.To reduce the complexity of equation solution,a fast method called split method is introduced.A multi-scale framework is established to perform the deblur procedure from large scale to small scale until the blur kernel is gotten.Finally,the estimated kernel function and clear image are used as the initial value for the next scale.Using the estimated kernel,the final clear image can be gotten by the total variation method.As compared with traditional image restoration methods,the proposed method can obtain the exact solution by direction solution of the energy equation conveniently without predetermining and selecting the image gradients.Experiments demonstrate the feasibility and validity of the proposed method.