ECG denoising and compression by sparse 2D separable transform with overcomplete mixed dictionaries

In this paper, an algorithm for ECG denoising and compression based on a sparse separable 2-dimensional transform for both complete and overcomplete dictionaries is studied. For overcomplete dictionary we have used the combination of two complete dictionaries. The experimental results obtained by the algorithm for both complete and overcomplete transforms are compared to soft thresholding (for denoising) and wavelet db9/7 (for compression). It is experimentally shown that the algorithm outperforms soft thresholding for about 4 dBor more and also outperforms Extended Kalman Smoother filtering for about 2 dB in higher input SNRs. The idea of the algorithm is also studied for ECG compression, however it does not result in better compression ratios than wavelet compression.