Denoising of ECG signal by non-local estimation of approximation coefficients in DWT

Abstract

This paper presents an ECG denoising technique using merits of discrete wavelet transform (DWT) and non-local means (NLM) estimation. The NLM-based approach is quite effective in removing low frequency noises but it suffers from the issue of under-averaging in the high-frequency QRS-complex region. In addition to that, the computational cost of NLM estimation is also high. The DWT, on the other hand, is effective in removing high-frequency noise but needs larger decomposition levels in order to denoise the low-frequency components. Thresholding lower-frequency components in the DWT domain often results in a loss of critical information. To overcome these drawbacks, in the proposed method, two-level DWT decomposition is first performed in order to decompose the noisy ECG signal into low- and high-frequency approximation and detail coefficients, respectively, at each level. The high frequency noise is removed by thresholding the detail coefficients at both the levels. The noise in the lower-frequency region is then removed by performing NLM estimation of Level 2 approximation coefficient. The Level 2 approximation coefficients actually represent the low-frequency envelope of the ECG. Thus, the proposed technique effectively combines the power of both NLM and DWT. At the same time, the computational cost of whole process is not more than the earlier existing techniques since NLM estimation is performed only on Level 2 approximation coefficients instead of the complete ECG signal. The proposed method is found to be superior to the existing state-of-the-art techniques when tested on the MIT-BIH arrhythmia database.