Send the link below via email or IMCopy
Present to your audienceStart remote presentation
- Invited audience members will follow you as you navigate and present
- People invited to a presentation do not need a Prezi account
- This link expires 10 minutes after you close the presentation
- A maximum of 30 users can follow your presentation
- Learn more about this feature in our knowledge base article
Arabic characters recognition neural networks
Transcript of Arabic characters recognition neural networks
•Presence of noise or salt-and-pepper in the scanned image.
•Linguistic problems. Handwritten Text Specific Difficulties •Variety in character size.
•Variety in writing instrument
•Different writers or even the same writer under different conditions may slant their letters differently.
Arabic Text Specific Difficulties •Arabic characters can have more than one shape according to their position in a word whether at the beginning, middle, final, or stand alone, as shown in Figure 1.
•Different writers or even the same writer under different conditions will write some Arabic characters in completely different ways, as shown in Figure 2. •Other characters have very similar contours and are
difficult to recognize especially when non-character
and external objects are present in the scanned
image. Figure 3 shows a list of such characters.
•Handwritten Arabic characters depend largely on contextual information. There are a lot of handwritten characters that can be classified into two or more different classes depending on whether you look at them separately, in a word, or even in a sentence. For example, (zero) in Arabic looks exactly
the same as a dot ‘.’.
Methods Perception net (natsha)
inputs are a Matrix of 50x50 Pixels
Each output neuron has a bias
Bias is the orginal value for neuron The total input to the jth neuron output neuron is calculated by:
Y=(∑WijXi)+ bi (1)
Where, bj is bias weight for the jth neuron. Weights are an given initial zero values
Wjnew= Wjold+ α (tj - yj )Xj (2)
bjnew= bjold+α (tj - yj ) (3)
α = Learning rate
tj and yj = Target and actual values for the jth output neuron, respectively.
The values for xi, yj and tj are either +1 or -1. Multilayer BP-NN A multilayer neural network using back propagation training policy.
- It consists an input layer, one hidden layer, and output layer.
-the input layer consists of 50x50 pixels
- the hidden layer consists of 20x20 pixels
The output layer consists of as many neurons as the required number of characters to be recognized.
The output layer has a bias weight The sigmoid function evaluated as the net output value
for that neuron, which can be expressed mathematically as:
σk= (tk – yk).f ́ (yk) (4)
Where, tk target output for the kth output neuron. Wik new =Wik old +α.σk .Zj
a is the learning rate, typically in the range of 0 to 1. Tests And Results Training and Testing times for the network Testing rates with no addition of noise Testing rates with addition of 10% noise Testing rates with addition of 20% noise Error Rates Presentation By:
Anas Al Hamdan Omar Al Smeheen Eyad Jbran
References 1- Back Propagation Neural Network Arabic Characters
Classification Module Utilizing Microsoft Word
Hamza, Ali A. Faculty of Computer Science and Information Technology, Al- Isra Private University. 2008
2- Arabic Text Recognition
Ramzi Haraty and Catherine Ghaddar, Lebanese American University, Lebanon 2004.
3- Arabic Character Recognition using Artificial
Neural Networks and Statistical Analysis
Ahmad M. Sarhan, and Omar I. Al Helalat, 2007 Thank You