1 / 20

Real-time Computer Vision with Scanning N-Tuple Grids

Real-time Computer Vision with Scanning N-Tuple Grids. Simon Lucas Computer Science Dept. Outline. Background: N-Tuple Classifiers The scanning n-tuple grid Isolated Character Recognition Isolated Face Recognition Convolutional Mode OCR Real time vision demo Conclusions.

gerda
Download Presentation

Real-time Computer Vision with Scanning N-Tuple Grids

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Real-time Computer Visionwith Scanning N-Tuple Grids Simon Lucas Computer Science Dept

  2. Outline • Background: N-Tuple Classifiers • The scanning n-tuple grid • Isolated Character Recognition • Isolated Face Recognition • Convolutional Mode OCR • Real time vision demo • Conclusions

  3. N-Tuple Classifiers • Work by randomly sampling input space • First applied to binary images • Very fast; reasonable accuracy • Scanning N-Tuple classifier (Lucas, 1995) • Applied to sequence recognition • Fast and accurate • Current work • SNT Grid • Specially developed for convolutional (sliding window) applications • Recognise patterns independent of location

  4. SNT-Grid System Architecture Binarise (e.g. Niblack) Scanning Index (SNT-Grid) Likelihood Image Likelihood Image Integrated Likelihoods Further Processing (e.g. Dictionary or Language Model)

  5. Original

  6. Binarised

  7. SNT Indexed

  8. Simple Operation • Slide grid over image • Interpret each position as binary number

  9. Efficient Implementation • Very simple idea • Decompose one 2-d scan • Into two 1-d scans! • Reduces time complexity • Suppose image is n x n • Window is m x m • Reduce from O(n2m2) • To O(n2) • Well worth the effort!

  10. Worked Example

  11. SNT Indexing: Java Code

  12. OCR Results: MNist Digits

  13. SNTGrid Speed on MNist • Java Implementation • Chars are 28 x 28 grey level images • Training (60,000 chars) • 8s (> 7,000 cps) • Testing (10,000 chars) • 3.8s (> 2,600 cps)

  14. ORL Face Data • 40 subjects • 10 images from each • Using 5 for training, 5 for testing • Average around 97.5% accuracy • Competitive with other methods • Much faster!

  15. Museum Archive Cards • Hard to read with conventional OCR

  16. 2 Detector : Raw outputs

  17. ‘2’ Detector – Integrated OP(Uses Integral Array of Viola + Jones)

  18. Real-time Demo • Very efficient • Can use it for real-time expression recognition • Or a ‘video’ joystick! • Bit like EyeToy – but potentially more sophisticated

  19. Sample testsReal-time Demo

  20. Conclusions • Basis of simple and efficient computer vision • Trick is the scan decomposition • Also use of integral image to accumulate likelihoods • Currently being applied to reading text in natural scenes • Many other applications also • Further reading: ICDAR 2005 Paper (on my web page)

More Related