激光与光电子学进展, 2018, 55 (11): 111005, 网络出版: 2019-08-14
基于多尺度的扫描电镜图像无参考质量评价方法 下载: 1055次
No-Reference Quality Assessment Method of Evaluating Scanning Electron Microscopy Images Based on Multi-Scale Characteristics
图像处理 图像质量评价 多尺度 扫描电镜 对比度失真 image processing image quality assessment multi-scale scanning electron microscopy contrast distortion
摘要
扫描电镜(SEM)图像能够直观地显示微观世界。在SEM成像过程中,需要反复调整设备参数来保证图像的对比度,这个过程费时费力。提出一种基于多尺度的SEM图像对比度失真的无参考质量评价方法,指导成像参数的选择。首先,建立了SEM图像数据库,进行主观实验,得到相应的主观平均意见分数(MOS);然后,根据人类的视觉系统具有多尺度特性,提取图像不同尺度奇异值分解域相似度、频域和熵共10个特征,结合MOS值通过支持向量回归训练回归模型;最后,利用前述模型预测图像的质量分数。实验结果表明,本文方法与主观评价结果保持很高的一致性,其性能优于主流的全参考和无参考质量评价方法。
Abstract
Scanning electron microscopy (SEM) imaging can visually reveal the microscopic world. In SEM imaging, the device parameters must be repeatedly adjusted to ensure the optimum image contrast. This process is often time-consuming and labor-intensive. We propose a novel no-reference quality assessment method for evaluating the SEM image contrast distortion based on multi-scale characteristics, which can be used as a guide to select imaging parameters. Firstly, a SEM image database is established, and the corresponding subjective mean opinion score (MOS) is obtained via subjective experiments. According to the multi-scale characteristics of the human visual system, 10 features are extracted, including singular value decomposition similarity with different scales, frequency domain features, and entropy. The MOS values and 10 features are then used to train a regression model via support vector regression. Finally, this model is used to predict the image quality score. The experimental results reveal that the proposed method can maintain a high level of consistency with subjective evaluation results, and its performance is superior to the mainstream full-reference and no-reference quality assessment methods.
李巧月, 商钢城, 田强, 陈曦, 韩习习, 周玉, 李雷达. 基于多尺度的扫描电镜图像无参考质量评价方法[J]. 激光与光电子学进展, 2018, 55(11): 111005. Qiaoyue Li, Gangcheng Shang, Qiang Tian, Xi Chen, Xixi Han, Yu Zhou, Leida Li. No-Reference Quality Assessment Method of Evaluating Scanning Electron Microscopy Images Based on Multi-Scale Characteristics[J]. Laser & Optoelectronics Progress, 2018, 55(11): 111005.