Wavelet based watermarked normal and abnormal heart sound identification using spectrogram analysis

• Published in: 2012 IEEE International Conference on Computational Intelligence and Computing Research pp. 1 - 7
• Main Authors: Dey, Nilanjan, Mishra, Girish, Nandi, Bijurika, Pal, Moumita, Das, Achintya, Chaudhuri, Sheli Sinha
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.12.2012
• Subjects: Authentication; Discrete wavelet transforms; Euclidean distance; Heart; Heart Sound; Random sequences; Spectrogram; Stationary Wavelet Transform (SWT); Training; Valves; Vectors; Watermarking; Wavelet Transform (DWT)
• ISBN: 1467313424; 9781467313421
• DOI: 10.1109/ICCIC.2012.6510173

Present work proposes a computer-aided identification of watermarked normal or abnormal heart sound based on Wavelet Transformation for tele-diagnosing of heart diseases. In this proposed method, the heart sound is converted into 2-D square matrix form. The 2-D signal is decomposed into four sub bands using Stationary Wavelet Transformation. HH 1 sub band is further decomposed using Stationary Wavelet Transformation. Watermark image is embedded within the generated HH 2 sub band. During embedding, watermark image is dispersed within the selected sub-band using a random sequence and a Session key. After embedding the watermark, that 2-D signal is taken as an input of spectrogram analysis. Due to the presence of Cumulative Frequency components in the spectrogram, DWT is applied on the spectrogram up to n level to extract the features from the individual approximation components. One dimensional feature vector is obtained by evaluating the Row Mean of the approximation component of the spectrogram. Watermark image is retrieved from the de-noised Watermarked Heart Test Sound. Instead of considering the heart sound samples, the present approach recognizes the set of spectrograms as the database. Thereafter 1-D feature vectors are attained by evaluating the Row Mean of the spectrogram approximation components obtained from the trained samples. Subsequently, minimum Euclidean distance between the Feature vector of the test heart sound and the Feature vectors of the trained heart sound samples is determined to identify the heart sound. By applying this algorithm, almost 78% of accuracy is achieved whereas without watermarking the accuracy is almost 82%.