ABSTRACT
The Non-Local Means (NLM) image de-noising algorithm uses the similar structures of the whole image to improve the signal-to-noise ratio of the output image. There are some shifting and rotating changes in the time-domain images captured by the same camera. In this paper, a layered three dimensional non-local means de-noising algorithm is proposed by utilizing these rotating and shifting similarities to further improve the quality of the output image. In the first layer, all images are de-noised by the NLM algorithm respectively. These de-noised images are used to form a three dimensional tower in the second layer, and the similarities between the reference block and the matched blocks within the search window are calculated based on the Euclidean distance. The adaptive filter parameter is determined by the Otsu’s method, and those matched blocks with small similarity values are excluded from the estimation of the output pixel. The experimental results show that our method can obtain better de-noising results as well as preserving more details.
