Fuyuan Mei1, Dong Zhang2, Yan Yang1. 1. School of Physics and Technology, Wuhan University, Wuhan, Hubei 430072, PR China. 2. School of Physics and Technology, Wuhan University, Wuhan, Hubei 430072, PR China. Electronic address: dongz@whu.edu.cn.
Abstract
BACKGROUND AND OBJECTIVE: In the observed medical ultrasound image, there is always some speckle noise which suppress the details of images and impairs the value of ultrasonography in diagnosis. This work present a novel despeckling method which effectively exploit non-local self-similarity for restoration of corrupted ultrasound images. The proposed approach consist of three stages. First, an improved optimized Bayesian non-local means (OBNLM) filter in which pixel patch is represented by a new vector form is used to get an preliminary estimation of noise-free image. Then, a new index called redundancy index of each pixel patch is calculated for determining which areas in image have low redundancy. Finally, another new vector form is used to represent pixel patch in areas with low redundancy obtained in second stage to recalculate filtered output, and the recalculated output is superimposed on preliminary estimation to generate final result of proposed method. METHODS: The performance of proposed approach is evaluated on simulated and real ultrasound images. The experiments conducted on various test image illustrate that our proposed algorithm outperforms the various classic denoising algorithms included block matching 3-D (BM3D) and optimized Bayesian non-local means filter. RESULTS: The objective evaluations and subjective visual inspection of denoised simulated and real ultrasound images demonstrate that the proposed algorithm can achieve superior performance than previously developed methods for speckle noise suppression. CONCLUSIONS: The combined use of two new representations improve denoising and edge preserving capability of proposed filter apparently. The success of proposed algorithm would help in building the lay foundation for inventing the despeckling algorithms that can make fuller use of information in images.
BACKGROUND AND OBJECTIVE: In the observed medical ultrasound image, there is always some speckle noise which suppress the details of images and impairs the value of ultrasonography in diagnosis. This work present a novel despeckling method which effectively exploit non-local self-similarity for restoration of corrupted ultrasound images. The proposed approach consist of three stages. First, an improved optimized Bayesian non-local means (OBNLM) filter in which pixel patch is represented by a new vector form is used to get an preliminary estimation of noise-free image. Then, a new index called redundancy index of each pixel patch is calculated for determining which areas in image have low redundancy. Finally, another new vector form is used to represent pixel patch in areas with low redundancy obtained in second stage to recalculate filtered output, and the recalculated output is superimposed on preliminary estimation to generate final result of proposed method. METHODS: The performance of proposed approach is evaluated on simulated and real ultrasound images. The experiments conducted on various test image illustrate that our proposed algorithm outperforms the various classic denoising algorithms included block matching 3-D (BM3D) and optimized Bayesian non-local means filter. RESULTS: The objective evaluations and subjective visual inspection of denoised simulated and real ultrasound images demonstrate that the proposed algorithm can achieve superior performance than previously developed methods for speckle noise suppression. CONCLUSIONS: The combined use of two new representations improve denoising and edge preserving capability of proposed filter apparently. The success of proposed algorithm would help in building the lay foundation for inventing the despeckling algorithms that can make fuller use of information in images.
Authors: Dominik Vilimek; Jan Kubicek; Milos Golian; Rene Jaros; Radana Kahankova; Pavla Hanzlikova; Daniel Barvik; Alice Krestanova; Marek Penhaker; Martin Cerny; Ondrej Prokop; Marek Buzga Journal: PLoS One Date: 2022-07-07 Impact factor: 3.752