Nishanth Lingamneni Program Manager Microsoft Corporation SYS-007T.
March 6 th, 2010 Khai Nguyen Grace Park Matthew Pham Nishanth Alapati Trevor Carothers Sky Lin...
-
Upload
nelson-caldwell -
Category
Documents
-
view
218 -
download
1
Transcript of March 6 th, 2010 Khai Nguyen Grace Park Matthew Pham Nishanth Alapati Trevor Carothers Sky Lin...
March 6th, 2010
Khai NguyenGrace ParkMatthew Pham
Nishanth AlapatiTrevor CarothersSky Lin
MENTOR: Jeff Wilhelm
Project Description How It Works Algorithm – SIFT Algorithm – Blob Detection Algorithm – Correlation Conclusion Demo Future Work
Create a mobile application that assists disabled people in identifying U.S. currency
The user will photograph bills and the application will say the denomination out loud
The user takes a picture of a dollar bill The application sends the picture to a
server The program on the server determines
the denomination of the bill The server returns the result to the
phone, which says the denomination of the bill That’s a
twenty20
VLFeat – an open source library developed by grad students at UCLA Vision Lab
SIFT detects keypoints from reference images
Descriptors uniquely identify keypoints
Keypoints of new images are compared to the keypoints of our reference images to find a match
Robust to changes in scale, rotation
Limitation: Can’t handle rotation. Can’t handle both side of the bill Can’t handle lighting effect of the bill
Constraints: requires the whole bill in view.
Sensitive to noise. Lack of resilience to rotation, scaling. Conclusion: Can’t compare to SIFT!
SIFT works best for currency recognition, due to its invariance to scale, rotation, and image quality.
Precompute SIFT descriptors corresponding to template images.
More testing to refine SIFT parameters!