Farabi Hasan Harija Yalamanchi
Farabi Hasan Harija Yalamanchi
Farabi Hasan Harija Yalamanchi
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
<strong>Farabi</strong> <strong>Hasan</strong><br />
<strong>Harija</strong> <strong>Yalamanchi</strong>
• Object Recognition<br />
• Face Recognition<br />
• Optical Character Recognition
Object Recognition using<br />
Eigenspace<br />
Template matching problem<br />
The idea is to represent image onto a<br />
lower dimensional space
Preliminaries<br />
Objects should be centered in the image<br />
Constant background<br />
Square matrix<br />
Illumination conditions
Algorithm<br />
1. Obtain training images I1, I2…IM for the database with<br />
different angles and different lighting conditions.<br />
2. Calculate the mean image<br />
3. Normalize each image by subtracting the mean image<br />
4. Create a matrix A = [Φ1 Φ2 . . . ΦM] where Φ is<br />
normalized image<br />
5. Compute first few eigenvectors of A as they contain the<br />
most important information<br />
6. Each normalized face Φ i in the training set can be<br />
represented as a linear combination of these<br />
eigenvectors
7. Each normalized training face Φi is represented in<br />
the basis by a vector<br />
8. Obtain the test image<br />
9. Subtract the mean image from the test image<br />
10. Compare the Euclidean distance and<br />
11. These distances show how far/close the test image is<br />
from the database of images
Results for Object Recognition<br />
Database of images used:
Distances for a cup:<br />
d1 = 296.4598<br />
d2 = 1.4176e+003<br />
d3 = 1.0786e+003<br />
d4 = 99.3106<br />
d5 = 1.2149e+003<br />
Distances for a box:<br />
d11 = 1.4838e+003<br />
d12 = 3.1979e+003<br />
d13 = 2.8589e+003<br />
d14 = 1.6810e+003<br />
d15 = 2.9952e+003
Very similar algorithm!<br />
Minor differences
Distances:<br />
d1 = 246.7174<br />
d2 = 806.5047<br />
d3 = 246.7174<br />
d4 = 5.0179e+003<br />
d5 = 5.1997e+003<br />
d6 = 4.9331e+003<br />
d7 = 1.3278e+003<br />
d8 = 1.3006e+003<br />
d9 = 1.3448e+003<br />
Distances:<br />
d1 = 2.9477e+003<br />
d2 = 2.3879e+003<br />
d3 = 2.9477e+003<br />
d4 = 1.8235e+003<br />
d5 = 2.0053e+003<br />
d6 = 1.7387e+003<br />
d7 = 4.5222e+003<br />
d8 = 4.4950e+003<br />
d9 = 4.5392e+003
Distances:<br />
d1 = 2.3897e+003<br />
d2 = 2.7448e+003<br />
d3 = 178.2691<br />
d4 = 314.1939<br />
d5 = 4.4594e+003<br />
Distances:<br />
d1 = 2.0783e+003<br />
d2 = 1.7233e+003<br />
d3 = 4.6463e+003<br />
d4 = 4.7822e+003<br />
d5 = 8.6497
Similar Algorithm
Database<br />
Each letter and number is represented as a .bmp<br />
file.<br />
…<br />
…
Drivers License<br />
Verify age of license holder
Steps to OCR<br />
1. Take picture of license<br />
2. Crop DOB<br />
3. Create edge view<br />
4. Clip the image horizontally<br />
5. Clip the segmented image vertically<br />
6. Use Matlab’s corr2() function to find the<br />
correlation of the current character to the<br />
characters in the database
Steps (cont…)<br />
7. Find the character in the database with the highest<br />
correlation to the current character and store the<br />
value<br />
8. Print the value in text file<br />
9. Verify Age
Practical Uses<br />
Using recognition<br />
Verify a person’s face from database<br />
Using OCR<br />
Verify that the person is of legal age to drink, smoke,<br />
etc..<br />
Identification checker and if camera available, verify<br />
the person on the license is the same person in real<br />
life<br />
Reducing Fake ID usage
References:<br />
http://www.cse.unr.edu/~bebis/CS791E/Notes/<br />
http://www.cs.ait.ac.th/~mdailey/matlab/<br />
http://cswww.essex.ac.uk/mv/allfaces/faces94.html<br />
http://www.ele.uri.edu/~hansenj/projects/ele585/OCR<br />
http://www.mathworks.com/matlabcentral/fileexchange/loadFile.do?<br />
objectId=18169&objectType=file