Improving dimensional measurement from noisy atomic force microscopy images by non-local means filtering

Quantitative evaluation of dimensional parameters from noisy atomic force microscopy (AFM) images was investigated. Non-local means (NLM) denoising was adopted to reduce noise and maintain fine image structures. Major tuning parameters in NLM filtering, such as the patch size and the window size, were optimized on simulated surface structures. The ability of dimensional evaluation from noisy data was demonstrated to be improved by almost 15 times. Finally, NLM filtering with optimal settings was applied on experimental AFM images, which were scanned on a patterned few-layer graphene specimen. Evaluations of the step height and the pattern size were verified to be much more accurate and robust. Such a data processing method can enhance the AFM dimensional measurements, particularly when the noise-level is reached.