You are on page 1of 12

Hand writing recognition by

artificial neural network


Introduction
• Hand writing recognition is the ability of computers
to detect and interpret the handwritten text coming
from an sources such as documents, touch screens
or other devices.
• This type of complex pattern recognition does not
give best performance with conventional computers.
• But nowadays new processing models called
artificial neural network is used for such kind of
application
Neural network
• A neural network consists of an interconnected
group of artificial network.
• These neural networks are dynamic in nature
means their functionality changes by adapting
weights at the interconnections.
• Basically neural network contain three types of
layer
1. Input layer
2. Hidden layer
3. Output layer
• Inputs to the neural network is handwritten text
coming from any devices.
• There must be some character database
• Output of network is recognize text
• Neural network uses some algorithm to match input
with is best suitable output
Back propagation algorithm

• Simplicity and reasonable speed makes back


propagation algorithm best suitable for most of
application.
• Back propagation trains network by adjusting
weights and biases.
• For accurate recognition we must provide
handwritten input in very large numbers by different
people.
Input-Output to the network
• Network does not recognize original character, so we have
to first scan written character and make image of it.
• Image of character contains number of pixels in each row
and column.
• No. of neuron in input layer is depend upon the no. of
pixels.
• For 10x10 image input layer should contain 100 neuron.
• Output layer contains 26 neuron for recognize different 26
layer.
Network structure
• For the accurate network it should train
with very large number of samples given
from different people because everyone has
different style of writing.
• Sometimes thinning and zooming of image
is also useful for feature extraction
Weakness
• Due to excessive training time consumption is
higher.
• One trained network is only useful for only one
language.
• Extra tools for character scanning, image cropping
,thinning and binarization is sometimes reqiured.
Applications

• Used in mobile phones and PDA


• For security purpose, i.e. authentication
• To make criminal records
Conclusion
Despite variations in character size, orientation, and
position, the neural network was still able to
recognize many of the characters. Different
customization is used to implement newer word or
character recognition. James Pittman and Mitica
Manu were the original inventor of the handwritten
recognition neural network.
References
• Hand written recognition by james mittman and
mitica mmanu , US patent 7428516B2
• Pooja singla, hand writing recognition system using
neural networks of international journel of
computing and application,Vol 5,june 2010
• Alexander J. Faaborg ,Using Neural Networks to
Create an Adaptive Character Recognition System,
Cornell University, Ithaca NY (May 14, 2002)

You might also like