Home My Page Chat Tik-Toke

Himanshu Mazumdar
Software Experience
Hardware Experience
Summary of Career
Selected Publications
Internet Links 
Scientific Projects
Neural Network Projects 
Firmware Projects
Software Projects
Full Papers 
Dr.Himanshu S. Mazumdar, Senior member IEEE

Neural Network Projects

·          Hand Written Text Recognition System for Visually Impaired

·          Neural Network Toolbox using C++

·          Simulation of Two Legged Robot

·          Prediction of Time Series Data

·          Pattern Recognition Toolbox

·         Teacher Classroom Scheduling Algorithm

Title: Hand Written Text Recognition System for Visually Impaired

Objective: To recognize hand written capital letter English character from a text line and produce the corresponding Speech Output to enable blind persons to read (listen) hand written notes.

  Description: A multi-layer neural network is used to recognize the hand written English characters. The net is trained with 10 sets of hand written English alphabets. The process involves (a) boundary detection, (b) thinning, (c) normalizing to 8x8 size (d) inputting to a multi-layer neural network and calculating the output value (e) selecting the best and converting to ASCII value (f) outputting pre-recorded speech corresponding the ASCII values.

  Funding Agency: DDIT, University, Nadiad , India


Title: Neural Network Toolbox using C++

Objective: To develop an interactive optimum Artificial Neural Network designer software capable of designing the network using GUI, interfacing to input-output and training the network to learn the input-output set.

  Description: A neural network toolbox is designed to graphically build the neural network   and use it for desired applications like modeling, prediction, mapping, pattern recognition etc. The toolbox supports ordered or random connectivity with generic learning algorithm and network optimizing tools like pruning. The toolbox supports different neuron activation functions. This toolbox is used different applications and algorithms. This work receiver the best paper award from Computer Society of India (CSI).

Funding Agency: Physical Research Laboratory, India


Title: Simulation of Two Legged Robot

Objective: To demonstrate autonomous walking of a human skeleton using a neural network based controller under computer simulation.

  Description: A ten joint human skeleton is simulated. In each joint independent controller are used. The controllers are inputted with center of gravity (CG) of the skeleton. Angular displacement of each joint is independently controlled using multi-layer neural networks and a rulebook. Rulebook is used to define walking.    

This type of robots can be used for autonomous planetary exploration.

  Funding Agency: Physical Research Laboratory, India


Title: Prediction of Time Series Data

Objective: To demonstrate the prediction complex of time series functions like day temperature, stock market, process variables, etc using time delay Neural Network.

  Description: A multiplayer neural network is used with some hidden neurons having previous samples memory retained while working on current sample. Such network is trained with current data as desired output and past data as input. In many cases it successfully learned the input output set.  The system is then given current data as input in many applications it predicated future output. Such network is found to build a model of input output function. System is successfully tested with complex periodic function.

Funding Agency: Physical Research Laboratory, India


Title: Pattern Recognition Toolbox

Objective: To develop a generic interactive toolbox to develop image pattern recognition applications.

  Description: A Neural Network based package is developed using VC++. The package has user-friendly GUI front end, multi-layer and Khoenen’s neural network as pattern recognition engine. The system supports other image processing tools.  

Funding Agency: DDIT, University, Nadiad, India


Title: Teacher Classroom Scheduling Algorithm

Objective: To develop neural network based algorithm to get optimum combination of teacher classroom scheduling. Teacher for a classroom is selected based on the subject skill level required for a class.

  Description: An algorithm is developed to solve the teacher classroom-scheduling problem using Neural Network. A potentially non-conventional type iterative learning strategy is developed that produces very close to optimum combination in few seconds.

  Funding Agency: DDIT, University, Nadiad, India