基于L0稀疏先验的相机抖动模糊图像盲复原

发布时间：2018-03-06 06:32

本文选题：模糊图像　切入点：图像盲复原　出处：《光学精密工程》2017年09期 　论文类型：期刊论文

【摘要】：提出了一种基于L0稀疏先验的改进正则化模糊图像盲复原算法来解决相机抖动所产生的模糊问题。根据模糊图像的梯度分布要比清晰图像稠密并且暗通道的稀疏性也相对较小这一固有属性建立了新的优化模型。针对L0范数的高度非凸性和暗通道稀疏优化过程中涉及到的非线性最小化问题,提出了一种近似线性映射矩阵,并用半二次分解法对L0最小化问题进行求解。最后,采用快速傅里叶变换在频域中对模糊核及清晰图像进行交替迭代运算得到复原图像。对多幅不同类型的模糊图像进行了实验,结果显示:复原图像平均灰度梯度高达11.411,图像信息熵达到7.304,处理365×285的图像只需8.07s。提出的算法有效抑制了图像边缘处的振铃效应,完整保留了清晰的细节信息的同时显著提高了运算速度,并适用于多种不同类型图像的盲复原。
[Abstract]:An improved regularized blur image restoration algorithm based on L0 sparse priori is proposed to solve the blur problem caused by camera jitter. The gradient distribution of blurred image is denser than that of clear image and the dark channel is sparse. A new optimization model is established for the inherent property of relatively small property. For the highly non-convexity of L _ 0 norm and the nonlinear minimization problem involved in the sparse optimization of dark channels, a new optimization model is proposed. In this paper, an approximate linear mapping matrix is proposed, and the semi-quadratic decomposition method is used to solve the L0 minimization problem. Fast Fourier transform (FFT) is used to iterate the fuzzy kernel and clear image in frequency domain to get the reconstructed image. The results show that the average grayscale gradient of the reconstructed image is 11.411, the information entropy of the reconstructed image is 7.304, and the processing of 365 脳 285 images requires only 8.07 s. The proposed algorithm can effectively suppress the ringing effect at the edge of the image. The algorithm not only preserves the clear detail information but also improves the speed of operation and is suitable for blind restoration of many different types of images.
【作者单位】：中国科学院长春光学精密机械与物理研究所;中国科学院大学;
【基金】：林业公益性行业科研专项基金资助项目(No.201204515)
【分类号】：TP391.41

【相似文献】