15-10-2012, 05:47 PM
Name of the Organisation
Name of the Organisation.docx (Size: 303.38 KB / Downloads: 27)
Abstract
Handwritten character recognition is considered as one important criteria of the image processing.Improving the accuracy of the recognition characters by reducing generalization problem by Multiscale training technique.
This project is mainly aims at the development and implementation of handwritten character recognition system for different resolution of images for surveillance applications.
Overview of hand written character recognition
Handwritten character recognition is a challenging task due to the unconstraint shape variations, different writing styles and different kind of noise that breaks the strokes in number or changes their topology.
Handwritten Character Recognition
Techniques for handwritten character recognition can be grouped into on-line and off-line recognition. Online recognition means that machine recognizes the writing while user writes. Off line recognition by contrast, is performed after writing is completed.
To differentiate between off-line and on-line handwriting recognition systems, the fundamental difference is in the nature of the handwriting sample used in each system. In off-line systems, the samples static. The system analyzes a digital image using either the raw pixel data, or some sort of representation of the pixel data only data that can be obtained directly from the image is used, there is no additional information given to it. This differs from an online system, where the sample analyzed in a dynamic environment. This allows on-line systems to collect detailed real time information. This information may include the pressure and speed that a sample is written with, and the specific order in which the different alphanumeric characters placed. In the project using an offline system for handwriting recognition.
There exist several different techniques for recognizing characters. Common technique uses back propagation in a neural network and how good a neural network solves the character recognition problem.
Components of Recognition
The Character Recognition System must first be created through a few simple steps in order to prepare for recognition. An image of each letter of an alphanumeric must be in the form of matrix. A character matrix is an array of black and white pixels, the vector of 1 represented by black, and 0 by white. Multiple fonts of the same alphabet may even be used under separate training sessions.
A digital images is an image f(x, y) refers to a two-dimensional intensity function f(x,y), where x and y denote co-ordinates of the image. A digital image can be considered as a matrix whose row and column indices identify a point in the image and the corresponding matrix element value identifies the gray level at that point. The elements of such array called “pixels”.
Image Acquisition
The process of acquiring image using different input modality is known as image acquisition. There several techniques to acquire images. Image can be taken through digital camera, electronics tablets, scanner, and paintbrush screen etc.
Input data sets required for this project are hand printed characters and handwritten characters. Hand printed characters acquired from the computers of different sizes and different styles. Handwritten characters prepared on a piece of blank paper by using pen to write down all of 62 characters. Samples of 5 of each character having different style and sizes of high and low resolutions of images were used, hence a total of 310 characters were used for recognition. 26 upper case letters and 26 lower case letters, with 10 numerical digits.
Binarization
In general, an image has three channels: Red, Green and Blue. These channels represents different nuance of the above color. Put one over the other, they form all the known colors. For example, Red=0, Green=0, Blue=0 represents black and Red=255, Green=255, Blue=255 represents white.
The basic way to create a grayscale image is to set the Red, Green, and Blue to the same value. The operation that converts a grayscale image into a binary image is known as binarization.
In this project, binarization process has done automatically by using thersholding method by neural network toolbox. Thersholding refers to setting all the gray levels below a certain level to zero or above a certain level to a maximum brightness level. Each pixel is assigned a new value (1 or 0).
Preprocessing
Preprocessing of handwritten data is done prior to the recognition stage. Preprocessing is an importance step before extraction of feature to train large, high-resolution nets. Network with a very large input and training by the usual back-propagation methods, not only is the training slow but also the generalization properties of such nets poor.
Given a character image as input, the features extractor derives the features that the character possesses. Some features that have been carried for character recognition geometric features, direction features. The derived features then be used as input to the character classifier. In our project the features input vectors which fed into the neural network for training and for recognition.
Resizing
The resize block enlarges or shrinks an image by resizing the image along one dimension (row or
column). Then, it resizes the image along the other dimension (column or row).
The character each of these images needs to be resized to form standard image dimensions.
Hence the images of each character resized into 3 dimensions of 20 by 28 pixels, 10 by 14 pixels, and 5 by 7 pixels for neural network training in order to perform multiscale technique.