SMITA VEMULAPALLI

Download: PDF   DOC 

144 Ponce de Leon Ave NE, Apt # 1214,
Atlanta, GA 30308.
Email: smita@ece.gatech.edu
Webpage: http://users.ece.gatech.edu/~smita/

Aug 2006 - present

Aug 2004 - May 2006

Aug 2000 - May 2004

Jan 2005 - present

Jun 2006 - Aug 2006

May 2003 - Jun 2003

Audio-Video Based Handwritten Mathematical Content Recognition For Classroom Videos
-Prof. Monson H. Hayes III, Georgia Tech, USA.

Aug 2004 - present

Our goal is to design and implement a handwritten mathematical content recognition system. Initially we worked on tablet PC input and later, during my summer internship at Texas Instruments E&PS, we also looked at other input devices that could be used for easier input of mathematical expressions into the calculator. We are currently using classroom videos as the input and have completed the pre-processing stage which involves the extraction of text written on the whiteboard from the foreground objects such as the instructor and also the background i.e. the whiteboard. Since mathematical content is not expressed using cursive handwriting, the character segmentation stage is quite simple. Our current approach to the character recognition problem involve the use of both audio as well as video to recognize the mathematical content written on the whiteboard. When there is some ambiguity in the recognition of a character written on the whiteboard, we will perform Hidden Markov Model(HMM) based speech recognition over a short window of speech to resolve the ambiguity. We are working on the speech recognizer which will recognize mathematical words i.e. numbers, alphabets, math operators and Greek symbols, from a dictionary of a few hundred words. We are also exploring alternate techniques such as word-spotting to overcome some of the problems in the HMM based speech recognizer due to mismatch in the grammar of the input and that of the recognizer. The bigger challenge would be to formulate a probabilistic technique by which the location of the word in the speech segment and the words before and after it would help in synchronizing the audio and video and therefore enable integration of speech information with the video. Decision making in case of conflicts between the video and speech recognizers would be resolved by using a combination of their confidence measures. If the errors in recognition of the symbols written on the whiteboard are considered to be independent of the errors in speech recognition of the same symbol, the recognition accuracy will be increased by using both the audio and video. This research also involves the analysis of the structure of equations by using logical and spatial relationships between recognized symbols and once again, we intend to use speech information to resolve ambiguity. Such an audio-video based approach would also prove to be useful for training the system for new users.

Implementation of Reconstruction from Projections Using Filtered Backprojections
-Prof. Russell M. Mersereau, Georgia Tech, USA.

Aug 2005 - Dec 2005

This project involved implementation of the filtered backprojection algorithm which reconstructs an image from its projections. The image was reconstructed using different values of ∆θ, different resolutions and also different filters in the filtering stage. Fourier Reconstruction of the image was also attempted.

Fricative Detection
-Prof. Chin-Hui Lee, Georgia Tech, USA.

Jan 2005 - Apr 2005

The goal of this project was to implement a fricative detector. In addition to implementing fricative detection, segmentation and classification was also implemented for certain other phonemes. This project could either be a part of a knowledge-based phoneme recognition system or be used as the front-end for a statistical speech recognition system, for example, one that uses HMMs or ANNs.

Implementation of Inte polation Schemes and Point-Based Registration for Medical Images
-Prof. Oskar Skrinjar, Georgia Tech, USA.

Jan 2005 - Apr 2005

Various interpolation schemes were implemented on a Magnetic Resonance Image (MRI) of the Human Brain. The original image was rotated 10 degrees and interpolation was performed to get the intensity values at original pixel positions. This rotation and interpolation was repeated till a complete 360 degree rotation was achieved and the final interpolated image was compared with the original image. The interpolation schemes implemented included Nearest Neighbor, Bilinear and Spline. Next, point-based image registration was implemented on MRI images of the Human Heart with known correspondences for Identity, Rigid Body, Affine and Thin Plate Spline transformations.

Real-Time Implementation of a Speaker Verification System
-Prof. P. T. Vanathi, PSG College of Technology, India.

Dec 2003 - May 2004

Our approach was a combination of stochastic and template modeling to implement a text-dependent system. Discrete HMMs were used to match sequences of vector-quantized mel-cepstral coefficients. We made use of the Bakis 5-state HMM and the Baum-Welch algorithm to generate probability measures. Codebook initialization and training was done using the LBG algorithm and the Lloyd?s algorithm respectively. The system was trained and tested using speech samples recorded in different environments.

-  Master of Ceremony for the National Conference on Signals, Systems and Security (NCSSS) 2002 and the National Conference on emerging trends in VLSI Design And Testing (NCVDAT) 2003 held at PSG College of Technology, Coimbatore, India.

-  Designed and implemented a circuit for an Automatic Plant Irrigation System which was displayed at the Golden Jubilee Exhibition held at PSG College of Technology, Coimbatore, India in July 2001.

-  Attended a six-week training program in LabVIEW in the National Instruments funded Virtual Instrumentation Laboratory at PSG College of Technology, Coimbatore, India.

-  Ranked among top 0.2% in the Tamil Nadu Professional Courses Entrance Examination 2000 conducted for admission to engineering colleges.