You are on page 1of 16

Project Members:

Mohammad Nazmul Haque


Roni Shikder
Susmita Baidya
Ehsan Uddin Ahmed

Project Presentation of
Bangla Digit Recognition using NN
Submitted to:
Prof. Syed Akhter Hossain, Ph.D
Daffodil International University
Presentation Outline
1. Image Preprocessing
 Image Acquisition
 Convert to Gray Scale
 Convert to Binary Image
 Identify maximum connected component
 Crop Individual character & make raw data
 Resize character into 5x7 ,applying fuzzy technique
2. Create Neural Network
3. Training the Network
4. Testing The Network
5. Performance Evaluation
1. Image Preprocessing
 Image Acquisition

 Convert to grayscale image


1. Image Preprocessing Contt.
 make the image binary image &check for the maximum
connected component
1. Image Preprocessing Contt..
 Crop individual characters & make raw data

 sub-images have been resize to 50 by 70


bw_7050=imresize(bw2,[70,50]);
 finding the average value in each 10 by
10 blocks
Atemp=sum(bw_7050((cnt*10-9:cnt*10),(cnt2*10-
9:cnt2*10)));
1. Image Preprocessing Contt..
image can be down to 5 by 7
matrices, with fuzzy value
for cnt=1:7
for cnt2=1:5
Atemp=sum(bw_7050((cnt*10-
9:cnt*10),(cnt2*10-9:cnt2*10)));
lett((cnt-1)*5+cnt2)=sum(Atemp);
end                    
                   
end                    
                   
 Resize Digit by 5x7 by average value          
           
 
 
 
 
 
 
 
 
 
                   
2. Create the NN
 Classification by simple FF-BP-NN
 The number of features, which is 5 by 7 = 35

 Performance Goal : 0.01


 Momentum constant : 0.95
 Epochs : 5000
3 . Training the Network

Training Output Training Performance

TRAINGDX, Epoch 0/5000, SSE 100.468/0.1, Gradient 51.6634/1e-006


TRAINGDX, Epoch 20/5000, SSE 39.1175/0.1, Gradient 0.755705/1e-006
TRAINGDX, Epoch 40/5000, SSE 39.7742/0.1, Gradient 0.247939/1e-006 3
Performance is 0.0846207, Goal is 0.1
TRAINGDX, Epoch 60/5000, SSE 39.8531/0.1, Gradient 0.1739/1e-006 10
TRAINGDX, Epoch 80/5000, SSE 39.8688/0.1, Gradient 0.159411/1e-006
TRAINGDX, Epoch 100/5000, SSE 39.8692/0.1, Gradient 0.160278/1e-006
TRAINGDX, Epoch 120/5000, SSE 39.8604/0.1, Gradient 0.171435/1e-006
2
TRAINGDX, Epoch 140/5000, SSE 39.827/0.1, Gradient 0.211754/1e-006 10
TRAINGDX, Epoch 160/5000, SSE 39.628/0.1, Gradient 0.438358/1e-006
TRAINGDX, Epoch 180/5000, SSE 34.2116/0.1, Gradient 2.91705/1e-006

Training-Blue Goal-Black
TRAINGDX, Epoch 200/5000, SSE 11.2205/0.1, Gradient 2.92171/1e-006
TRAINGDX, Epoch 219/5000, SSE 0.0846207/0.1, Gradient 0.160478/1e- 1
006 10
TRAINGDX, Performance goal met

0
10

-1
10

-2
10
0 20 40 60 80 100 120 140 160 180 200
219 Epochs
4. Testing the NN
 Testing Input:

 Testing Output:
1 3 10 4 10 7 6 8 9 5
 Testing Result Concatenated Individual Digits:
Performance : Case 1
Performance : Case 2
Performance : Case 3
Performance : Case 4
Performance Summary
Particulars Result
No of Training Digit 160
No of Testing Digit 40
No of Miss-Classification 2
Miss-Classified Digits ১, ৩ as ৯,০
Performance (Success Rate) 95%
Future Scope
 Create NN based Bangla Number recognition System
 Car’s Bangla Number Plate Recognition System
 Road Speed Limit recognition system for Bangla Road
Sign
Thank You
 Any Questions / Suggestions ?

You might also like