联合小波域和频域的图像去模糊算法

Joint Image Deblurring In Wavelet and Frequency Domain

提出了一种在小波域和频域上联合恢复模糊图像的算法。首先在小波域上对模糊图像去噪,提出按照贝叶斯公式估计出小波系数的收缩因子,恢复出模糊图像的小波系数值。此后,按照正则化反卷积图像恢复算法,对去噪模糊图像进行恢复。该算法使得反卷积时的正则化算子选取为较小的值,从而恢复的图像既滤除了噪声,同时降低了边缘模糊等振铃效应。实验结果表明,选择拉普拉斯正则化算子,该算法恢复的图像质量优于频域正则化反卷积算法,此外在同等噪声水平下,不同图像的最优正则化参数处在较小的相同动态范围之内,避免了恢复算法中的反复经验试值寻求最优。

A joint image deblurring algorithm in the wavelet and frequency domain is presented. First, the wavelet coefficients of the corrupted image shrink with the coefficients that are estimated by bayes equation. Therefore, we restore the blured image by this denoising method. Then, we deconvolve the denoised image with the reguralization. In the described algorithm, the optimal regularization parameter selects the low value. So, the restored image not only filter the noise, but also has minimal ringing. Experiments show that the quality of restored image with this algorithm by selecting the laplacian operator is superior to that of the regularized deconvolving method in the frequency. Morever, it avoids the optimal regularized parameter selected by experimental trying because the optimal parameter locates in the same narrow bound for different image in one noise level.