1 / 25

Recognition of Video Text Through Temporal Integration

Recognition of Video Text Through Temporal Integration. Trung Quy Phan , Palaiahnakote Shivakumara Tong Lu and Chew Lim Tan. Introduction. Text extraction from video frames  video search and retrieval. Introduction. Low resolution Complex background Unconstrained appearance.

lydie
Download Presentation

Recognition of Video Text Through Temporal Integration

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. Recognition of Video TextThrough Temporal Integration TrungQuyPhan, PalaiahnakoteShivakumaraTong Lu and Chew Lim Tan

  2. Introduction • Text extraction from video frames video search and retrieval

  3. Introduction • Low resolution • Complex background • Unconstrained appearance

  4. Introduction • Low resolution • Complex background • Unconstrained appearance • Temporal information

  5. Problem • Input • Word bounding box in a reference frame • Frame ID • Output • Binarized image • Scope • Static texts • Linearly moving texts

  6. Approach • Tracking • Alignment • Integration • Refinement

  7. 1. Tracking • Find • [tstart, tend]  text framespan • Bounding box in each frame  text instance tstart … … tend tref

  8. 1. Tracking • Text descriptors

  9. 1. Tracking • Text descriptors • Stroke Width Transform-SIFT

  10. 1. Tracking • t = tref + 1, tref + 2, … • Initialize search area

  11. 1. Tracking • t = tref + 1, tref + 2, … • Initialize search area • If matchRatio ≥ 0.1  estimate new BB

  12. 1. Tracking • t = tref + 1, tref + 2, … • Initialize search area • If matchRatio ≥ 0.1  estimate new BB • Otherwise, found tend

  13. 2. Alignment • Align at pixel-level  better integration

  14. 2. Alignment • Align at pixel-level  better integration • Slide reference text mask over individual masks  optimal alignment

  15. 2. Alignment • Align at pixel-level  better integration • Slide reference text mask over individual masks  optimal alignment

  16. 3. Integration • Text probability map

  17. 3. Integration • Initial binarization

  18. 4. Refinement • SWT: rounded strokes • Intensity values preserve sharp edges & holes suppress background pixels

  19. Experiments • Moving text dataset: English + German • 250 words • 1,545 characters • Bottom to top, right to left and left to right • Static text dataset: English • 212 words • 1,389 characters

  20. Experiments • Methods for comparison • Niblack (Single) • Min/max (Multiple) • Average-Min/max (Multiple) • Ours (Single) • Ours (Multiple)

  21. Sample Results

  22. Sample Results

  23. Results on Moving Texts • Character recognition rate (CRR) • Word recognition rate (WRR)

  24. Results on Static Texts • Multiple-frame: ~20% improvement over single-frame

  25. Summary • A variation of SIFTfor robust tracking • Integration based onword masks • Future work: handle complex text movements

More Related