For learning how to write and pronounce English characters

as opposed to on-line recognition where the computer recognizes the characters as they are drawn.About ocr OCR is the acronym for Optical Character Recognition.  Optical recognition is performed off-line after the writing or printing has been completed.  Optical character recognition is needed when the information should be readable both to humans and to a machine  Both hand written and printed characters may be recognized  It converts scanned images of machine-printed or handwritten text (numerals.Introduction 1. . and symbols) into a computer-processable format. letters.

FreeTTS is JSAPI speech synthesis engine that we have used . About speech synthesis  The text-to-speech (TTS) synthesis procedure consists of two main phases.2. . text analysis (input text->Phonetic o/p)and speech generation(phonetic info->acoustic o/p)  Conversion of text into Speech can be implemented using java Speech Application Programming Interface (JSAPI) through which applications can use functionality of speech engines.

compare the scanned character with a pre-defined shape. • Speech synthesis has reached a high level of performance. but there is still much improvements to be done to achieve more natural sounding speech. These readers scan the shape of a character on a document. and high intelligibility in synthesis. and convert the character into its corresponding bit pattern for storage in main computer memory. This technology is still in development.Current status of development • OCR readers can convert typed and handwritten documents into digital data. with low error rates in text analysis. .

.advantages     reduces the time required by user to enter the data. no requirement of keyboard for entering text . Helps in learning language along with spoken help.  people with reading disabilities(dyslexics). can use it.  Both writing and Pronounciation can be learned. during his learning phase as many times as required.  a person can change his own handwritten pattern of alphabet (improved) . A computer with handwriting recognition integrated with speech synthesis can teach any time at any place.

. The objective of the neural network is to transform the inputs into meaningful outputs.Artificial neural networks Output Inputs An artificial neural network is composed of many artificial neurons that are linked together according to a specific network architecture.

The normalization factor is the reciprocal of the square root of the vector length.Kohonen algo  The input to a Kohonen neural network is given using the input neurons. One output neuron is chosen as the winner  To examine which neuron would win and produce output. These input neurons are each given the floating point numbers that make up the input pattern to the network. This is done by summing the squares of the input vector. steps to be followed are :  Normalize the inputFirst calculate the "vector length" of the input data. Then. . A Kohonen neural network requires that these inputs be normalized to the range between -1 and 1. determine the normalization factor.

Contd…  Calculate each output neuron’s o/p – For each of the output neurons . the output neuron that has the largest o/p value becomes the winner.  This o/p must now be normalized by multiplying it by the normalization factor  The above calculated o/p is mapped to bipolar number by adding 1 and dividing the result by 2.  Finally. . choose the winning neuron . dot product of input vector and connection weights between the input neurons and that output neuron must be calculated.

Unsupervised learning  No help from the outside  no information available on the desired output  Learning by doing .

 Cropping is done. Training the network to recognize same or identical patterns. is used for character recognition and training.  the bit pattern of the image is grabbed. an algorithm to reduce the resolution of the letters being drawn. for eliminating the extra white space around     the image. Recognition (using Kohonen Self Organization Map) and speech synthesis (Using JSAPI).(by classifying to the same output neuron) Error calculation(how well network classifies) . DownSampling.Processes in our OCR…  The hand written characters are first drawn using the mouse.

the original Abstract Windows Toolkit (AWT) and the newer Swing Swing components have the prefix J to distinguish them from the original AWT ones (eg JFrame instead of Frame).Language used-Java  JAVA is a general computer programming language      developed by Sun Microsystems. java.awt.swing. . Object oriented language platform independent code written in JAVA will be easier to maintain and reuse in the long run Java has two GUI packages. To include Swing components and methods in your project you must import the java.*.* packages.*. and javax.event.awt.

synthesis. setVisible(bool).y.w. The setDefaultCloseOperation(constant) method controls the action that occurs when the close icon is clicked.  JPanel is the most commonly used content pane.util . The add() method allows GUI components to be added to the pane. setSize(w.  For text-to-speech conversion using java we need some packages eg. maximize.h). setResizable(bool). It adds basic functionality such as minimize. An instance of the pane is created and then added to a frame. isResizable() and getTitle(). Containers are used to hold and group components such as text fields and checkboxes etc.h). close. title and border to basic frames and windows. JFRAME AND JPANEL  JFrame is the most commonly used top-level container.freetts etc and some jar files to be installed in our working folder before compiling our program. Some important JFrame methods are: setBounds(x.speech. The way they are added is controlled by the current layout manager. . setTitle(str).

Data flow diagrams .

Dfd .

Level-2 Data Flow Diagram and written haracters Dfd User interfac e Croppin g DownSampl ed image Kohonen Neural Network Input vecto r Connecti on Weights Outputs Vector Length Normalize d input Normaliz ed Outputs Recogniti .

so that most of our characters can be recognized. Epoch occurs when training data is presented to the network .features in ocr  It can recognize handwritten characters and simultaneously speak that recognized character. which can be loaded and the application can be trained further to recognize characters drawn by us .error is calculated and weights are adjusted to reduce error. The process for training a Kohonen neural network involves stepping through several epochs until the error of the Kohonen neural network is below acceptable level. .  We can train the network to recognize our own handwriting .  We have a training file that contains training samples for our own handwriting (capital versions of 26 english letters ).

lastError.for the epoch that just occurred.. The second error. by selecting a particular letter. indicates the best(least) lastError that has occurred so far. bestError.during training the n/w.  It can create a list of letters that the program has been trained for.  Error i. of times we have tried to adjust weight matrix to reduce error . how well the training inputs(the letters that you created)  The first error.we can retrain our program for that letter. indicates the total error for the Kohonen neural network . Tries counts the no. there is no requirement of any further training .Features cond…. . and deleting it . map to the output neurons(26 characters).e.If the error is below the acceptable level of error(10%).

Snapshot .

 than .

Sign up to vote on this title
UsefulNot useful