Medical Intelligence and Language Engineering Lab    TTS Demo   |   Downloads   |   Videos   |   Contact Us   |   Site Map      
     Home   |    About Mile   |    Projects   |    Research Area   |    Publications   |    Alumni   |    FAQ's    |    News & Events   |    Gallery

Research Area
Current Projects
Publications
Download Area
  

Our Past Research Areas:

3D Image Registration and Compression

Analysis and 3-D Compression of MRI Images

Development of algorithms for 3-D volume measurement of each of the organs in the brain from 2-D slices obtained from MRI. This is for comparing MRI images from untreated schizophrenia patients with those from normal subjects. This work is planned in collaboration with Prof. P. N. Jayakumar, Head of Neuroradiology, and Prof. B. N. Gangadhar, Dept. of Psychiatry, NIMHANS, Bangalore. We have received the first set of transverse section, T1 & T2-weighted images in the DICOM format and started analyzing them. Also planned is the 3-D compression of set of more than 100 slices of the brain.

Estimation of Neurological Signals

Evoked potentials (EP) signals are the electrical responses of the brain time-locked to an external stimulus. The SNR of the EP signals is typically very low as they are severely corrupted by the background spontaneous brain activity. Conventionally few hundreds of responses obtained by applying identical stimuli are averaged in order to estimate the underlying evoked response. The focus of our research is towards reduction in the experimental time or to obtain a meaningful signal estimate using fewer number of responses. We investigate non-linear filtering approaches in the wavelet domain for evoked potential signal estimation. The methods proposed exploit inter-scale correlation property of the undecimated wavelet transform of the noisy observations. Yet another approach involves using Gaussian radial basis functions for approximating the underlying evoked signal.

Fetal Lung Maturity Analysis

Ultrasound Image Processing: Fetal lung and liver tissues were examined by Ultrasound (US) in 240 subjects during 24 to 38 weeks of gestational age, in order to determine the relationship between the gestational age and the textural features of sonograms of fetal lung. Since the histology of the liver remains constant with respect to gestational age, features from the lung region are compared with those of liver. Fractal features, features obtained from spatial gray level dependence matrices, energy measures from Laws' textural masks and statistical features derived from the histogram are used. The features do not unambiguously determine the maturity of the fetal lung. Effective compression was achieved for sector scan US images, using Hough transform for segmentation, and VQ & run-length coding for compression of the gray-scale and text regions.

Research in the following areas of Biometrics had been carried out -

1. Speaker identification systems – using features such as pitch contour, area ratio contour, spectral and/or cepstral signatures.
2. Signature verification systems - from both static signatures and dynamic information from graphic tablets, using different neural architectures.
3. Face recognition systems, using Gabor space-scale features.

Speaker Recognition

Speaker recognition is the process of establishing the identity of a person from his/her voice. In speaker recognition, the voice is recognized as belonging to one speaker from a known set of speakers. In speaker verification, the problem is to decide, if the speaker is the person whom he/she claims to be. In both cases, the system can be text-dependent, text-prompted or text-independent. Features that have been used for speaker recognition are LP Coefficients, LP- Cepstrum, LP-dCepstrum, Mel-Cepstrum, Mel-dSepstrum, and Filterbank log energies. Some of the methods used for speaker modeling are Vector Quantization, Gaussian Mixture Model and Hidden Markov Model


© 2010 Medical Intelligence and Language Engineering Lab - IISc Campus, Bangalore.