1 / 81

Multi-scale Image Harmonization

Multi-scale Image Harmonization. Image Compositing. luxa.org. Current approaches. Alpha Matting Combine pixel values to create blended boundaries [ Porter & Duff ’84, Smith & Blinn ’96, … ]. Chuang et al. ’01. Current approaches. Alpha Matting

dung
Download Presentation

Multi-scale Image Harmonization

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. Multi-scale Image Harmonization

  2. Image Compositing luxa.org

  3. Current approaches • Alpha Matting • Combine pixel values to create blended boundaries • [ Porter & Duff ’84, Smith & Blinn ’96, … ] Chuang et al. ’01

  4. Current approaches • Alpha Matting • Combine pixel values to create blended boundaries • [ Porter & Duff ’84, Smith & Blinn ’96, … ] • Gradient-domain fusion • Combine pixel gradients to create seamless boundaries • [ Pérez et al. ’03, Agarwala et al. ‘04, … ] Pérez et al. ’03

  5. What if the images are incompatible?

  6. Copy & paste

  7. Gradient-domain fusion

  8. Our result

  9. Gradient-domain fusion Our result

  10. Given two images, “harmonize”them (i.e. match their appearance), and then composite them (with plausible boundaries).

  11. Multi-scale Image Harmonization • Harmonization • Color • Contrast • Noise • Texture • Blur • Compositing • Matte-based boundaries • Seamless boundaries

  12. Overview

  13. Overview

  14. Overview

  15. Overview

  16. Overview

  17. Overview

  18. Overview

  19. Overview

  20. Starstock / Photoshot Source Target

  21. Build pyramids Source Target

  22. * f1 * f2 … Image * fn Undecimatedpyramid subbands Analysis filters [ Burt ‘81, Burt & Adelson ’83 ]

  23. Build pyramids Source Target Source subbands Target subbands

  24. Source Target Source subbands Source subband histograms Target subband histograms Target subbands

  25. Source Target Histogram Matching Source subbands Source subband histograms Target subband histograms Target subbands [ Heeger & Bergen ’95 ]

  26. Source

  27. Histogram matching

  28. Over-sharpening Haloing Histogram matching

  29. Source Target Histogram Matching Source subbands Source subband histograms Target subband histograms Target subbands

  30. Histogram Matching as a Gain function Source subband "Harmonized" subband Histogram Matching Target subband

  31. Histogram Matching as a Gain function Source subband "Harmonized" subband Histogram Matching "Harmonized" subband values Source subband values

  32. Histogram Matching as a Gain function Source subband "Harmonized" subband Histogram Matching "Harmonized" "Harmonized" subband values Gain = Source Source subband values Source subband values

  33. Histogram Matching as a Gain function Source subband "Harmonized" subband Gain function "Harmonized" "Harmonized" subband values Gain = Source Source subband values Source subband values

  34. Histogram Matching Source Source subbands "Harmonized" subbands Gain functions

  35. Histogram Matching Source Problem: Arbitrary gain functions distort subbands. Source subbands "Harmonized" subbands Gain functions

  36. Smooth Histogram Matching Source Solution: Control gain functions to minimize distortion. Source subbands "Harmonized" subbands Gain functions

  37. Smooth Histogram Matching • Avoid large values in the gain functions. Source subband "Harmonized" subband Gain function

  38. Smooth Histogram Matching • Avoid large values in the gain functions. • Ensure that effect of gain functions is spatially smooth. [ Li et al. ’05 ] Source subband "Harmonized" subband Gain function

  39. Histogram Matching

  40. Smooth Histogram Matching

  41. Smooth variation No haloing Histogram Matching Smooth Histogram Matching

  42. Overview

  43. Overview

  44. Pyramid compositing = fi * Final composite Analysis Filters (used to create pyramids) Harmonized pyramid ?

  45. Boundary constraints Matte-based boundary Seamless boundary Chuang et al. ’01

  46. Pyramid compositing = fi and * Final composite Analysis filters Harmonized pyramid Boundary constraints ? ? • Can apply alpha matte-based boundaries, • seamless boundaries, or a combination of both

  47. Pyramid compositing Solve a sparse linear system of equations for composite = fi and * Final composite Analysis filters Harmonized pyramid Boundary constraints ? ? Linear filters Linear constraints

  48. Harmonized source pasted onto target

  49. Pyramid compositing with seamless boundaries

  50. Source Target Our composite

More Related