| 摘要: |
| 目的 一类特殊的图像在数字系统中扮演着重要的角色,如文本、条形码和指纹图像,这类图像的主要特征
是它们的像素值只能在一个二值集合中取值。 针对含有加性噪声和运动模糊的此类图像的恢复问题,提出一种新
的基于即插即用框架下的盲图像云运动模糊算法。 方法 在盲去模糊模型中结合了去噪器先验和二值先验知识,它
在利用先进去噪器( 如数据驱动的去噪器) 的同时使恢复的潜在图像具备二值特征。 通过提高潜在图像的恢复质
量,进一步改善估计模糊核的准确性,从而提升最终的恢复结果。 此外,为了进一步增强恢复图像的像素值分布特
性,通过对恢复的图像应用阈值化操作,限制恢复图像的像素值分布在特定的二值集合中。 结果 大量的数值实验
结果表明:该方法在处理含有噪声的运动模糊二值图像的任务中具有较好的应用效果,其性能优于现有的传统算
法。 结论 结合去噪器先验和二值先验知识的盲去模糊算法能够有效地恢复含加性噪声的运动模糊二值图像。 |
| 关键词: 二值图像 图像盲去模糊 即插即用 舍入算子 |
| DOI： |
| 分类号: |
| 基金项目: |
|
| Blind Binary Image Deblurring Using Plug-and-Play Algorithm |
|
YANG Xuesong HE Liangtian
|
|
School of Mathematical Science Anhui University Hefei 230601 China
|
| Abstract: |
| Objective A special class of images such as text barcodes and fingerprint images plays a significant role in
digital systems. The main characteristic of these images is that their pixel values can only take values from a binary set. To
address the problem of recovering degraded binary images containing additive noise and motion blur a new blind image
deblurring algorithm based on a plug-and-play framework was proposed. Methods This method combined denoiser prior
and binary prior knowledge in the blind deblurring model which utilizes advanced denoisers e. g. data-driven
denoisers while equipping the recovered latent images with binary features. By improving the recovery quality of the
latent image the accuracy of the estimated blur kernel was further improved which in turn enhanced the final recovery
results. In addition in order to further enhance the pixel-value distribution characteristics of the recovered images a
thresholding operation was applied to the recovered images to restrict the pixel-value distribution of the recovered images to
a specific set of binary values. Results The results of a large number of numerical experiments showed that this method
has good application in the task of processing blurred binary images with noise and motion blur and its performance is
better than the existing traditional algorithms. Conclusion Therefore the blind deblurring algorithm combining denoiser
prior and binary prior knowledge can effectively restore binary images degraded by additive noise and motion blur. |
| Key words: binary image blind image deblurring plug-and-play rounding operator |
