Automatic Logo Replacement
Upcoming SlideShare
Loading in...5
×

Like this? Share it with your network

Share

Automatic Logo Replacement

  • 836 views
Uploaded on

As a part of Class project for CIS 581 Sibashish and I presented this method to do automatic logo replacement.

As a part of Class project for CIS 581 Sibashish and I presented this method to do automatic logo replacement.

  • Full Name Full Name Comment goes here.
    Are you sure you want to
    Your message goes here
    Be the first to comment
    Be the first to like this
No Downloads

Views

Total Views
836
On Slideshare
834
From Embeds
2
Number of Embeds
1

Actions

Shares
Downloads
2
Comments
0
Likes
0

Embeds 2

http://www.slideshare.net 2

Report content

Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
    No notes for slide

Transcript

  • 1. Automatic Logo Replacement Sibasish Acharya and Saurabh Palan
  • 2. Project Overview
    • Dataset Creation
    • Feature Detection
    • Pair-wise logo alignment
    • Logo Similarity Measurement
    • Logo Warping
    • Logo Detection
    • Logo replacement
    • Logo appearance matching
    • Image Blending
    • Conclusion
  • 3. Dataset Creation
  • 4. Feature Detection
    • SIFT
      • Good but Not Panacea
      • Does not extract ample features
      • The clipart and reference logo lack similarity
    • HSV + PCA
    • Histogram of Orientation of Gradient (HOG)
      • Counts occurrences of gradient orientation in localized portions of an image
      • Nine bins
      • Extracts more features, thus comparative effective then SIFT
  • 5. Pair-wise logo alignment
    • RANSAC + TPS
      • Initial feature matches fed to RANSAC are detected by minimum SSD matching
      • 1000 iterations to Minimizes Bending Energy
      • In our case since the clipart is completely different from Logo, we select the best from the worst possible combination.
  • 6. Logo Similarity Measurement
    • Histogram comparison by  2 Difference
    • L1 Norm, L2 Norm
  • 7. Logo Warping
    • We obtained the best match clipart and warped it to the Logo
    • We use the parameters obtained from RANSAC + TPS for warping
  • 8. Logo Detection
    • Sliding Window
      • Obtain Patch
      • Compare with training Dataset
      • Compute dissimilarity with training dataset
      • Find minimum dissimilarity
        • If (Minimum dissimilarity <= threshold)
        • LOGO DETECTED
        • Else
          • NOT DETECTED
    • SVM
      • Did not work well…
  • 9. Logo replacement
    • Warp detected Logo towards reference logo
    • Warp replacement Logo towards detected logo
  • 10. Logo appearance matching
    • We use V of HSV
    • Calculate mean of V value of reference logo
    • Calculate mean of V value of detected logo
    • Add the differences to the logo clipart
  • 11. Image Blending
    • We calculate the bounding box from test image to be cropped out then replace bounding box by replacement logo using 3 level Pyramid Blending
  • 12. Conclusion
    • Efficiency of Feature Detection algorithm determined the outcome of the project…best algorithm for our application - HOG