Poisson Image Editing - CGL @ ETHZ CGL slideset 1 Poisson Image Editing Patric Perez, Michel...

download Poisson Image Editing - CGL @ ETHZ CGL slideset 1 Poisson Image Editing Patric Perez, Michel Gangnet,

of 31

  • date post

    25-Jun-2020
  • Category

    Documents

  • view

    0
  • download

    0

Embed Size (px)

Transcript of Poisson Image Editing - CGL @ ETHZ CGL slideset 1 Poisson Image Editing Patric Perez, Michel...

  • CGL slideset

    1

    Poisson Image Editing

    Patric Perez, Michel Gangnet, and Andrew Black (SIGGRAPH 2003)

    Seminar Talk by

    Tim Weyrich

  • CGL slideset

    2

    Overview

    • Guided Image Interpolation • Discretized Solution • Editing Operations • Discussion

  • CGL slideset

    3

    Interpolation Problem

    • f *: known image values • f : unknown values over region Ω • Assuming scalar image values

  • CGL slideset

    4

    Simple Interpolation

    • Maximize smoothness

    • Boundary constraints

  • CGL slideset

    5

    Simple Interpolation

    • Solution: Laplace Equation with Dirichlet boundary conditions

    • Membrane solution • Unsatisfactory due to over-blurring

  • CGL slideset

    6

    Guided Interpolation

    • v: guided field • v may be gradient of a function g

  • CGL slideset

    7

    Guided Interpolation

    • Minimize difference of gradient fields

    • Solution: Poisson Equation with Dirichlet boundary conditions

  • CGL slideset

    8

    Discrete Poisson Solver • Discretize directly by

    for neighbors p and q with

  • CGL slideset

    9

    Discrete Poisson Solver • Minimum satisfies linear system of equations

    If neighborhood overlaps boundary:

    For interior points:

  • CGL slideset

    10

    Discrete Poisson Solver

    • Linear system of equations – sparse (banded) – symmetric – positive-definite

    • Irregular shape of boundary requires general solver, such as – Gauss-Seidel iteration – Multi-grid

    • System can be solved at interactive rates

  • CGL slideset

    11

    Seamless Cloning

    • Importing Gradients from a Source Image g

    • Discretize

  • CGL slideset

    12

    Seamless Cloning Results

  • CGL slideset

    13

  • CGL slideset

    14

  • CGL slideset

    15

    Mixing Gradients • Two Variants

    – v averaged from source and destination gradients transparency

    – Select stronger one from source and destination gradients:

    Discretization:

  • CGL slideset

    16

    Mixing Gradients Results

  • CGL slideset

    17

    Mixing Gradients Results

  • CGL slideset

    18

    Mixing Gradients Results

  • CGL slideset

    19

    Texture Flattening

    • Preserve only salient gradients

    with masking function so that

  • CGL slideset

    20

    Texture Flattening

  • CGL slideset

    21

    Local Illumination Changes

    • Approximate tone mapping transformation after Fattal et al. 2002:

    • Attenuating large gradients

  • CGL slideset

    22

  • CGL slideset

    23

    Local Color Changes

    • Mix two differently colored version of original image – One provides f * outside – One provides g inside

  • CGL slideset

    24

    Local Color Changes

  • CGL slideset

    25

    Seamless Tiling • Select original image as g • Boundary condition:

    – f *north = f *south = 0.5 (gnorth + gsouth) – Similarly for the east and west

  • CGL slideset

    26

    Discussion

    Pros • Very general framework • No parameter tuning required • Method does not require precise selection • Versatile method

    – Seamless cloning, mixing gradients – Texture flatening – Local changes of illumination and color – Seamless tiling

  • CGL slideset

    27

    Discussion

    Cons • Cloning requires either of the images to be

    smooth • No refined selection is returned • Minimization only adapts low-frequency

    content – Potential color shift / re-coloring difficult to control – Dissatisfactory tiling – Cloning requires careful placement of prominent

    features

  • CGL slideset

    28

    Outlook

    • More image editing operators – Combinations (insert while flattening) – Other non-linear operations on gradients – More than one source images

    • Poisson editing of triangle meshes – Feature transfer – Detail preserving deformations

    • Other editing domains possible?

  • CGL slideset

    29

    Thanks

  • CGL slideset

    30

    Seamless Tiling

  • CGL slideset

    31

    Mixing Gradients Results