Docstoc

Inpainting Algorithms for Image Coding and Scaling Patrick Ndjiki Nya

Document Sample
Inpainting Algorithms for Image Coding and Scaling Patrick Ndjiki Nya Powered By Docstoc
					 Inpainting Algorithms for
Image Coding and Scaling
         Patrick Ndjiki-Nya
                               Outline
 Motivation
 Image Inpainting and Coding
   • Structure Interpolation
   • Post-Processing Methods
   • Results

 Image Inpainting and Spatial Scaling
   • Proposed Method
   • Results

 Conclusions and Future Work




                                                                                                                slide 2
                    P. Ndjiki-Nya – Inpainting Algorithms for Image Coding and Scaling @ ITG FA 3.2 12/7/2007
                           Motivation

High-frequency textures are difficult to code, e.g. flower bed,
crushed stones, water etc.




              Where is the original texture ?




                                                                                                                 slide 3
                     P. Ndjiki-Nya – Inpainting Algorithms for Image Coding and Scaling @ ITG FA 3.2 12/7/2007
                  Coding: Block Diagram


                                      MSE Data
Video In           Encoder                                   Decoder                               Video Out



                                 Synth. Metadata



                TAS                                         TS


           Texture Analysis-Synthesis (TAS) Video Coding




                                                                                                                    slide 4
                        P. Ndjiki-Nya – Inpainting Algorithms for Image Coding and Scaling @ ITG FA 3.2 12/7/2007
Coding: Synthesizer Input

             Current frame (t0)




                                                                                                 slide 5
     P. Ndjiki-Nya – Inpainting Algorithms for Image Coding and Scaling @ ITG FA 3.2 12/7/2007
     Coding: Rigid Texture Synthesizer (TSI)

  Left ref. frame (t0-x)            Current frame (t0)                              Right ref. frame (t0+x)




 Texture synthesizer for rigid (flower bed, sand, …) textures
 Parametric approach
 Features some similarity with global motion compensation
 Typical drawback: Limited by assumptions of motion model




                                                                                                                       slide 6
                           P. Ndjiki-Nya – Inpainting Algorithms for Image Coding and Scaling @ ITG FA 3.2 12/7/2007
  Coding: Non-Rigid Texture Synth. (TSII, 2D)

         Orig. texture                                                 Keep as is


                                                        Overlap




                    Patch #2   Patch #1                                          Synth. texture



                                        Stuffing procedure



 Generic approach: Good for rigid (flower bed, sand, …) as well as
non-rigid textures (water, smoke,…)
 Non-parametric approach




                                                                                                                            slide 7
 Typical drawback: Complexity of texture representation

                                P. Ndjiki-Nya – Inpainting Algorithms for Image Coding and Scaling @ ITG FA 3.2 12/7/2007
Coding: Non-Rigid Texture Synth. (TSII, 2D+t)




                                                                                                           slide 8
               P. Ndjiki-Nya – Inpainting Algorithms for Image Coding and Scaling @ ITG FA 3.2 12/7/2007
Coding: Intermediate Results




                                                                                                   slide 9
       P. Ndjiki-Nya – Inpainting Algorithms for Image Coding and Scaling @ ITG FA 3.2 12/7/2007
Coding: Intermediate Results (cont.)




                                                                                                      slide 10
          P. Ndjiki-Nya – Inpainting Algorithms for Image Coding and Scaling @ ITG FA 3.2 12/7/2007
Coding: Challenges




                                                                                              slide 11
  P. Ndjiki-Nya – Inpainting Algorithms for Image Coding and Scaling @ ITG FA 3.2 12/7/2007
Coding: Structure Interpolation




                                                                                                   slide 12
       P. Ndjiki-Nya – Inpainting Algorithms for Image Coding and Scaling @ ITG FA 3.2 12/7/2007
Coding: Structure Interpolation




                                                                                                   slide 13
       P. Ndjiki-Nya – Inpainting Algorithms for Image Coding and Scaling @ ITG FA 3.2 12/7/2007
Coding: Structure Interpolation Results




                                                                                                        slide 14
            P. Ndjiki-Nya – Inpainting Algorithms for Image Coding and Scaling @ ITG FA 3.2 12/7/2007
Coding: Structure Interpolation Results




                                                                                                        slide 15
            P. Ndjiki-Nya – Inpainting Algorithms for Image Coding and Scaling @ ITG FA 3.2 12/7/2007
Coding: Structure Interpolation Results




                                                                                                        slide 16
            P. Ndjiki-Nya – Inpainting Algorithms for Image Coding and Scaling @ ITG FA 3.2 12/7/2007
Post-Processing Methods: Poisson Cloning




                                                                                                                  slide 17
     Simple Cloning                                            Poisson Seamless Cloning

                      P. Ndjiki-Nya – Inpainting Algorithms for Image Coding and Scaling @ ITG FA 3.2 12/7/2007
Post-Processing Methods: Formalization

                                                       S                                  f*
                                                                                          f


                                                                    Ω

  v                                                                            ∂Ω
                  g



      • S: a closed subset of R2
      • Ω: a closed subset of S, with boundary ∂Ω
      • f*: known scalar function over S\Ω
      • f: unknown scalar function over Ω
      • ∂Ω: boundary pixels
      • g: source function




                                                                                                             slide 18
      • v: vector field of g

                 P. Ndjiki-Nya – Inpainting Algorithms for Image Coding and Scaling @ ITG FA 3.2 12/7/2007
Post-Processing Methods: Formalization

 Poisson equation with Dirichlet boundary condition

                                2
         min ∫ ∇ - v
                f                       with           f |∂Ω f * |∂Ω
                                                           =
          f   Ω


          Δ f = divv         over Ω with              f |∂Ω f * |∂Ω
                                                          =




                                                                                                              slide 19
                  P. Ndjiki-Nya – Inpainting Algorithms for Image Coding and Scaling @ ITG FA 3.2 12/7/2007
Coding: Structure Interpolation Results




                                                                                                        slide 20
            P. Ndjiki-Nya – Inpainting Algorithms for Image Coding and Scaling @ ITG FA 3.2 12/7/2007
Image Inpainting and Small Holes




                                                                                                     slide 21
         P. Ndjiki-Nya – Inpainting Algorithms for Image Coding and Scaling @ ITG FA 3.2 12/7/2007
                 Scaling Results




                                                                                                              slide 22
Bicubic Interpolation                           Structure-aware Interpolation

                  P. Ndjiki-Nya – Inpainting Algorithms for Image Coding and Scaling @ ITG FA 3.2 12/7/2007
          Conclusions and Future Work
 Efficient tools for image inpainting presented
 Performance optimization of cloning in inpainting
  framework
 Extension of structure-aware methods to video (rigid and
  non-rigid textures)




                                                                                                                slide 23
                    P. Ndjiki-Nya – Inpainting Algorithms for Image Coding and Scaling @ ITG FA 3.2 12/7/2007
Thank You




                                                                                               slide 24
   P. Ndjiki-Nya – Inpainting Algorithms for Image Coding and Scaling @ ITG FA 3.2 12/7/2007

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:2
posted:10/18/2012
language:English
pages:24