Abstract:
In this thesis we investigate wavelet transform based ECG compressiontechniques and compare them with conventional approaches. A major issue addressed ishow to guarantee a user-specified error limit measured by the percent root mean square difference (PRD) for the reconstructed ECG signal to be controlled at every segmentwhile keeping the compression ratio (CR) as large as possible with reasonableimplementation complexity.Two wavelet transform based compression methods, one based on discreteorthonormal wavelet transform (DOWT) and the other based on wavelet packettransforms are studied in detail. Decomposition, uniform quantization, and entropycoding are applied successively to compress the digital ECG signal while entropydecoding, and inverse transformation are applied to reconstruct the original signal.Different types of wavelet families are used to analyze the effect on CR and PRD. More conventional discrete sine / cosine transform based methods are also studied forcomparison purposes.Two numerical metrics PRD and CR are used as the major performanceevaluation parameters to quantitatively compare one method to another. The CR is ameasure of compression efficiency; the PRD gives information about the performance of the compression scheme and the distortion measured. Using the techniquesdeveloped, two different types of ECG signals (normal and an arrhythmic) arecompressed analyzed and the results are reported. In each technique, while the PRDincreases, the CR also increases. In general, the highest CR values are obtained with the wavelet transform; the lowest PRD values are obtained with the wavelet packettransform.|Keywords: Biomedical Signal Compression, Electrocardiogram, Wavelet Transform, Discrete Sine Transform, Discrete Cosine Transform, Arrhythmia.