George Hodgson Douglas Krofcheck Jeremy Leakakos Elizabeth Leib - - PowerPoint PPT Presentation

george hodgson douglas krofcheck jeremy leakakos
SMART_READER_LITE
LIVE PREVIEW

George Hodgson Douglas Krofcheck Jeremy Leakakos Elizabeth Leib - - PowerPoint PPT Presentation

Team Haplocyonopsis George Hodgson Douglas Krofcheck Jeremy Leakakos Elizabeth Leib David McClelland You'll have a list of 3 million books. You'll be given an image of a book, and you need to find out which of the 3 million it is -- in 9


slide-1
SLIDE 1

George Hodgson Douglas Krofcheck Jeremy Leakakos Elizabeth Leib David McClelland

Team Haplocyonopsis

slide-2
SLIDE 2

 You'll have a list of 3 million books. You'll be given

an image of a book, and you need to find out which

  • f the 3 million it is -- in 9 seconds.

 Performance requirements

  • Speed of identification
  • Accuracy of identification

 Identification simulation

slide-3
SLIDE 3

 Preprocessing Library

  • Precalculate various image values to allow for faster image

identification

 Identification Library

  • Algorithms using the preprocessed data to identify books

quickly and accurately

 Test Application

  • Application to demonstrate the use of the libraries and

allow for easy performance and accuracy testing

slide-4
SLIDE 4
slide-5
SLIDE 5
  • 1. Take picture
  • 2. Preprocessing
  • 3. Set pruning
  • 4. Identification
  • 5. Result
slide-6
SLIDE 6
slide-7
SLIDE 7

 Lack of test data

  • Utilized collection of images from Amazon

 Sponsor contact change

  • Complete project review

 Unfamiliar domain

  • Gathered knowledge
slide-8
SLIDE 8

 Libraries  Test application  Testing

  • Medium scale “perfect” samples

▪ 99.9% success rate

  • Small scale “realistic” samples

▪ 81.6% success rate

slide-9
SLIDE 9

 Algorithm documentation  System diagrams  Test application guide  Product implementation guide  Requirements specification

slide-10
SLIDE 10

 Improve existing algorithms

  • Higher identification rate
  • Faster identification

 Additional algorithms

  • Optical character recognition (OCR)
  • Wave transformations

 Large scale dataset  Automatic deployment

slide-11
SLIDE 11