1 / 8

Background Removal

David Harwin Adviser: Petros Faloutsos. Background Removal. Background Removal Process. Background Removal Process – An Example. Side Note. In this implementation, the segmenter is set to produce binarized masks corresponding to per-pixel FG/BG segmentation.

kipp
Télécharger la présentation

Background Removal

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. David Harwin Adviser: Petros Faloutsos Background Removal

  2. Background Removal Process

  3. Background Removal Process – An Example

  4. Side Note • In this implementation, the segmenter is set to produce binarized masks corresponding to per-pixel FG/BG segmentation. • Interestingly enough, the non-binarized grayscale difference has potential in background completion

  5. Measuring Accuracy • Raw accuracy (% correct pixels) show how well the removal was performed, but are dependent on the number of foreground pixels • Since this is essentially a classification problem, it is reasonable to define optimal behavior as minimizing both the false accept rate (FAR) and false reject rate (FRR)‏

  6. Sample results and conclusions

  7. Challenges and Ideas • A significant number of pixels either washed out or faded to black, making color differencing problematic • this suggests that black/white pixels should be treated as a special case • no success getting the framework to import video files, however • same method- frames read as still images • video formats not suitible for verification against ground truth data

  8. The next steps • current masks have rough edges and holes • filling algorithms largely domain-specific • smoothing – create averaged map at 1:2^x scale • try edge-finding algorithm and filling techniques • multiframe implementation – supplement BG model with motion likelihood updated each frame

More Related