Image denoising filter based on patch-based difference refinement

Sang Wook Park, Moon Gi Kang

Research output: Contribution to journalArticlepeer-review

4 Citations (Scopus)


In the denoising literature, research based on the nonlocal means (NLM) filter has been done and there have been many variations and improvements regarding weight function and parameter optimization. Here, a NLM filter with patch-based difference (PBD) refinement is presented. PBD refinement, which is the weighted average of the PBD values, is performed with respect to the difference images of all the locations in a refinement kernel. With refined and denoised PBD values, pattern adaptive smoothing threshold and noise suppressed NLM filter weights are calculated. Owing to the refinement of the PBD values, the patterns are divided into flat regions and texture regions by comparing the sorted values in the PBD domain to the threshold value including the noise standard deviation. Then, two different smoothing thresholds are utilized for each region denoising, respectively, and the NLM filter is applied finally. Experimental results of the proposed scheme are shown in comparison with several state-of-the-arts NLM based denoising methods.

Original languageEnglish
Article number067007
JournalOptical Engineering
Issue number6
Publication statusPublished - 2012 Jun

Bibliographical note

Funding Information:
This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MEST) (No. 2012R1A2A4A01003732).

All Science Journal Classification (ASJC) codes

  • Atomic and Molecular Physics, and Optics
  • Engineering(all)


Dive into the research topics of 'Image denoising filter based on patch-based difference refinement'. Together they form a unique fingerprint.

Cite this