ICISE23 - Biometric Recognition on Compressed ECG Signals Under Different Physiological Conditions

In the study, ECG is investigated as a biometric recognition system. To fit in the real-life scenarios, different physiological conditions and signal compression are considered. A total of 9 subjects conducting three simple exercises which are riding bicycle on high resistance, low resistance and walking taken from Physionet. Due to the noise collected during ECG recording process, the contaminated signals are filtered using low pass Butterworth filter. Then, QRS is segmented using Pan Tompkins algorithm. After that, Wavelet Transform is applied to compress the segmented QRS. Classification is done on non-compressed and compressed QRS by using multi-class of SVM to compare its accuracy. The study discovered that, biometric can be performed on compressed signals. The outcome also shows that, Cubic SVM is the proper classification technique to be applied on person verification regardless of different physiological conditions. The accuracy achieved was up to 94.4% for non-compressed signals and 91.7% for compressed signals.