Premium
Modified pyramid dual tree direction filter‐based image denoising via curvature scale and nonlocal mean multigrade remnant filter
Author(s) -
Teng Lin,
Li Hang,
Yin Shoulin
Publication year - 2017
Publication title -
international journal of communication systems
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.344
H-Index - 49
eISSN - 1099-1131
pISSN - 1074-5351
DOI - 10.1002/dac.3486
Subject(s) - artificial intelligence , noise reduction , computer science , pyramid (geometry) , bilateral filter , filter (signal processing) , computer vision , pattern recognition (psychology) , noise (video) , non local means , tree (set theory) , mathematics , image (mathematics) , image denoising , geometry , mathematical analysis
Summary To alleviate the disadvantage of traditional image denoising method in big images data, we propose a modified pyramid dual tree direction filter with nonlocal mean multigrade remnant filter for image denoising in this paper. The proposed denoising method is partitioned into 4 processes. Firstly, curvature scale model is used for building pyramid dual tree direction filter coefficients of noised image. Additionally, the coefficients are calculated by robust Bayes least square method. Then, we use pyramid dual tree direction filter inverse transformation to reconstruct an initial denoised image. At last, nonlocal mean multigrade remnant filter is adopted to filter the initial denoised image and we obtain the final denoised image. The proposed method completely used the multiscale and multidirectional selectivity with approximately translation invariance of pyramid dual tree direction filter. Finally, we assess the image denoising performance of the proposed approach over several test images and compare our results with the state‐of‐the‐art denoising algorithms. Our experiments show that the proposed image denoising method achieves better results than other methods. Furthermore, our new method not only effectively removes the noise but also better keeps the edge and detail information of texture and structure.