A Novel Hardware Architecture for Non-local Means Adaptive Filter
- PDF / 297,390 Bytes
- 4 Pages / 595.276 x 790.866 pts Page_size
- 57 Downloads / 223 Views
SHORT COMMUNICATION
A Novel Hardware Architecture for Non-local Means Adaptive Filter Nagapuri Srinivas1 • Pratik Singh1 • Gayadhar Pradhan1
Received: 18 May 2018 / Revised: 12 December 2018 / Accepted: 18 January 2020 Ó The National Academy of Sciences, India 2020
Abstract A novel dedicated hardware architecture is proposed in this letter for one-dimensional non-local means (NLM) algorithm. The patch-based NLM algorithm has been effectively used to filter out noises from the twodimensional images as well as various one-dimensional signals, such as an electrocardiogram (ECG) and audio signals. Dedicated hardware ensures the availability, predictability and stability of an algorithm. In this letter, an efficient hardware architecture of the NLM is proposed for removing noises from the corrupted ECG signal. The proposed architecture is implemented on a field-programmable gate array. The efficacy of the proposed architecture is verified by a firm comparison of the outputs obtained from both the hardware architecture and the software simulation performed on MATLAB. The computational complexity has also been compared for both cases (hardware and software) to show the effectiveness of the proposed hardware architecture. Keywords Non-local means Electrocardiogram Noise filtering Hardware architecture FPGA
In the recent years, the non-local means (NLM) algorithm has been successfully applied to filter out noises from & Gayadhar Pradhan [email protected] Nagapuri Srinivas [email protected] Pratik Singh [email protected] 1
Department of Electronics and Communication Engineering, National Institute of Technology Patna, Patna, India
electrocardiogram (ECG) and speech signals [1–7]. The patch-based NLM technique was originally proposed for enhancing the quality of noisy images [8, 9]. The performance of NLM adaptive filter is compared with other wellknown ECG noise filtering techniques based on discrete wavelet transform (DWT) and empirical mode decomposition (EMD) methods [1–3, 10–12]. The NLM filtering provides competitive performance compared to the stateof-the-art DWT- and EMD-based ECG noise filtering methods. NLM algorithm has firm mathematical background when compared to EMD-based methods which are empirical in nature. Also, NLM is only a single-level filtering technique, whereas DWT- and EMD-based methods use multi-level decomposition for noise filtering task. Motivated by these results, a novel dedicated hardware architecture for one-dimensional (1-D) NLM algorithm is proposed in this article. In NLM algorithm, for each sample point in the given noisy signal s(n), an estimate sd ðnÞ is computed as the weighted sum of the signal values at another noisy sample point s(m). The weight values are computed with the help of two local patches with starting point n and m, respectively. Both the patches consist of P number of sample points, and they lie within a predefined search-neighborhood N(n). The estimated filtered signal is computed as follows [1]: 1 X sd ðnÞ ¼ wðn; mÞsðmÞ ð1Þ WðnÞ mNðnÞ
The NLM estimation
Data Loading...
