Biometric identification using analysis of cardiac sound

Abstract

Human Heart Sound is unique in nature. It helps to regulate the pumping blood to the rest of the organ system for proper function, so that that pumping blood abruptly passes through the heart chamber to create heart sounds which are sounds as LUB and DUB via closure of Bicuspid and Tricuspid valve. These sounds having two segments S1 belongs to first sound and S2 belongs to second sound. In my works, first we made data collection from our ten volunteer of the age group 20-40 during three months period using Digital Stethoscope. We are having 100 heart samples stored in database. Then feature extraction using LFBC (linear frequency band cepstral), feature extraction method includes STDFT for converting the time domain signal into frequency domain. Then magnitude was taken and rejecting the phase part which generally include noise interference. Next the filter bank is applied, which reject the unwanted high frequency components. After that Dimension compression technique was used. Using DCT (Discrete Cosine Transform) here logarithmic first 24 coefficient was taken. Then Spike removal is done for removing the artifacts of position of hand movement while taking heart sound. At last, cepstral means subtraction is done, which removes the artifacts, here position of stethoscope is not same at all the time, after this operation is done, cepstral coefficient as our feature vector. Then Classification is done, using BP-MLP-ANN where 50 numbers of heart sound signal as Training and 50 numbers of heart sound signal as Testing are applied. The identification results show 52 % of performance accuracy.