针对数字图像在处理过程中容易产生模糊的现象, 提出了基于无参考图像质量评价的自适应反卷积去模糊算法。首先, 根据无参考图像质量评价结果与其失真等级的强相关性, 通过计算模糊图像的无参考评价参数确定图像的模糊等级, 进而根据图像模糊等级与模糊核的对应关系确定反卷积核; 其次, 提出将失真图像颜色空间转变到 YUV, 仅对失真图像 Y通道进行去模糊处理, 保证了彩色图像处理前后颜色的忠实性, 并提高算法运算效率; 最后, 针对图像灰度剧烈变化的邻域出现类吉布斯(Gibbs)振荡分布的现象, 提出基于梯度的权重矩阵进行控制。实验结果表明, 本文提出的算法在 Tid2008图库不仅能够对图像模糊进行快速有效去除, 并且恢复图像的纹理细节能够得到有效保留。

In view of the fact that digital images are prone to blur during processing, an adaptive deconvolution deblurring algorithm based on non-reference image quality evaluation is proposed. Firstly, according to the strong correlation between the no-reference image quality evaluation result and its distortion level, we can determine the fuzzy level of the image by calculating the no-reference image quality evaluation value and finally determine the convolution kernel with the linear relationship between the image fuzzy level and the fuzzy kernel. In order to ensure the fidelity of the color image before and after color processing and improve the efficiency of the algorithm, we propose to transform the distorted image color space to YUV, and only process the Y-channel in the distorted image. The Gibbs-like oscillation distribution phenomenon occurs in the neighborhood of sharp changes in the image gray levels. Gradient-based weight matrix is proposed to control the phenomenon. Experimental results show that the proposed algorithm can not only quickly and effectively remove the image blur, but also effectively retain the texture details of the restored image.