1 / 22

Scene illumination and surface albedo recovery via L1-norm total variation minimization

Scene illumination and surface albedo recovery via L1-norm total variation minimization. Hong-Ming Chen hc2599@columbia.edu Advised by: John Wright . Decomposition of a scene . =. .*. scene. illumination. Reflectance ( albedo ). .* : Matlab element multiplication operation.

basil
Download Presentation

Scene illumination and surface albedo recovery via L1-norm total variation minimization

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. Scene illumination and surface albedo recovery via L1-norm total variation minimization Hong-Ming Chen hc2599@columbia.edu Advised by: John Wright

  2. Decomposition of a scene = .* scene illumination Reflectance (albedo) .* : Matlab element multiplication operation

  3. Image Formation Pixel i = .* signals integration illumination reflectance scene intensity response Sensor response Light source power spectrum Object reflectance Sensor response (camera or eyes) :shutter speed, aperture size, quantization factor etc

  4. It is VERY HARD to directly model / simulate / solve this problem!

  5. Narrowing down our target problem • Simplification: • mean wavelength response (impulse response) • Assumption (on surface reflectance): • Lambertian Surface (Perfect diffuse reflection, no specular light) • Simulation (of light source model): • We need a formula to describe the behavior of the light source • Blackbody radiation: parameterize the light source with: • Light color (color temperature) • Light intensity

  6. Problem formulation:

  7. Assume: λRλG λG are known If there are N pixels in an image: 3Nobservations 5N unknowns (I, T, ref ) + 3 quantize factors log underdetermined system!

  8. Recovering unknown x • Previous approach • Introducing regularization terms into objective function • Current approach • Minimizing L1-norm total variation

  9. Previous Approach A result of: Intrinsic images by entropy minimization , Finlayson, ECCV2004 s p 1-D grayscale visualization 255 ps A segmentation-like result pp 0

  10. Drawbacks of this approach • There are at least 2 parameters (λ, σ) to be fine tuned. • The results of Finlayson’s approach heavily affects the accurateness of our prior. • 1. Its Achilles heel: projection problem • 2. it is still an open problem to find the best rotation angle.

  11. (λ =50 , σ = 10) (λ =10 , σ = 30) (λ =120 , σ = 5) (λ =120 , σ = 8)

  12. A brief review of Finlayson’ solution Its Achilles heel:

  13. L1 norm Total Variation Minimization Image From Wikipedia

  14. L1 norm Total Variation Minimization • Widely used in image denoise / Compressive sensing • E(x, y) + λTV(y). Image From Wikipedia

  15. Current approach: L1 TV norm • Applying L1-norm total variation on albedo term, • The L1-norm encourages a spiky result on gradient • Which means: we want most of the albedo gradients are 0 unless necessary => when albedo changes

  16. Results Original image Albedo (reflectance) image Light intensity image Light color (temperature) image

  17. Results Original image Albedo (reflectance) image Light intensity image Light color (temperature) image

  18. Results Original image Albedo (reflectance) image Light color (temperature) image

  19. Results Original image Albedo (reflectance) image Light color (temperature) image

  20. Editing Average T = 3940 Average T-1000 Original image Average T+1000 Average T+2000 Average T+3000 Average T+4000

  21. THANK YOU

More Related