35
Perceptually-Driven Statistical Texture Modeling Eero Simoncelli Howard Hughes Medical Institute, and New York University Javier Portilla University of Granada, Spain

Eero Simoncelli Howard Hughes Medical Institute, and New

  • Upload
    others

  • View
    2

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Eero Simoncelli Howard Hughes Medical Institute, and New

Perceptually-Driven Statistical Texture Modeling

Eero SimoncelliHoward Hughes Medical Institute, and

New York University

Javier PortillaUniversity of Granada, Spain

Page 2: Eero Simoncelli Howard Hughes Medical Institute, and New

What is “Visual Texture”?

Homogeneous, with repeated structures....

1/02

Page 3: Eero Simoncelli Howard Hughes Medical Institute, and New

What is “Visual Texture”?

Homogeneous, with repeated structures....

“You know it when you see it”

1/02

Page 4: Eero Simoncelli Howard Hughes Medical Institute, and New

Perceptual Texture Description

All Images

Texture Images

Equivalence class (visually indistinguishable)

Perceptual model:

• Set of texture images divided into equivalence classes (metamers)

• Perceptual “distance” between classes

1/02

Page 5: Eero Simoncelli Howard Hughes Medical Institute, and New

Julesz’s Conjecture (1962)

Hypothesis: two textures with identical Nth-order pixel statistics look the same(for some N).

• Explicit goal of capturing perceptual definition with a statistical model

• Statistical measurements should be:

– universal (for all textures)

– stationary (translation-invariant)

– a minimal set (necessary and sufficient)

• Julesz (and others) constructed counter-examples for N=2 and N=3, dis-missing the hypothesis...

1/02

Page 6: Eero Simoncelli Howard Hughes Medical Institute, and New

Julesz’s Conjecture, Revisited

Why did the early attempts fail?

• Right hypothesis, wrong model: A set of measurements equivalent to thevisual processes used for texture perception should satisfy the hypothesis.

• Lacked a powerful methodology for testing whether a model satisfies thehypothesis

• We can benefit from advances of the past few decades:

– scientific: better understanding of early vision

– engineering/mathematical: “wavelets”, statistical estimation, statisticalsampling

– technological: availability of powerful computers, digital images

1/02

Page 7: Eero Simoncelli Howard Hughes Medical Institute, and New

Testing a Texture Model

• As with most scientific test, we seek counter-examples

• Fundamental problem: we usually work with a small number of examples(tens or hundreds).

• Classification is an important application, but a weak test

• Synthesis can provide a much stronger test...

1/02

Page 8: Eero Simoncelli Howard Hughes Medical Institute, and New

Testing a Model via Synthesis

Example�����ture

Image

Random

Seed

Statistical

Image

Sampler

Statistical

Parameter

Estimator

Perceptual

Comparison

• Positive results are compelling, assuming:

– reference texture set contains a sufficient variety– statistical sampler generates “typical” examples

• Negative results are definitive: A single failure indicates insufficiency ofconstraints!

• Partial necessity test: remove a constraint and find a failure example

• Studying failures allows us to refine the model

1/02

Page 9: Eero Simoncelli Howard Hughes Medical Institute, and New

Methodological Ingredients

1. Representative set of example texture images: Brodatz, VisTex, our own

2. Method of estimating parameters: sample mean

3. Method of generating sample images from model: primary topic of thiswork

4. Perceptual test: informal viewing

1/02

Page 10: Eero Simoncelli Howard Hughes Medical Institute, and New

Iterative Synthesis Algorithm

Synthesis

Analysis

Transform MeasureStatistics

ExampleTexture

RandomSeed

SynthesizedTexture

Transform

MeasureStatistics

Adjust InverseTransform

Heeger & Bergen, ’95

1/02

Page 11: Eero Simoncelli Howard Hughes Medical Institute, and New

Transform: Steerable Pyramid

Example basis function Spectra

Linear basis: multi-scale, oriented, complex.

Basis functions are oriented bandpass filters, related by translation, dilation,rotation (directional derivatives, order K−1).

Tight frame, 4K/3 overcompleteness for K orientations.

Translation-invariant, rotation-invariant.

Motivation: image processing, computer vision, biological vision.

1/02

Page 12: Eero Simoncelli Howard Hughes Medical Institute, and New

Steerable Pyramid: Example Decomposition

Real part of coefficients complex magnitude of coefficients

Decomposition of a “disk” image

1/02

Page 13: Eero Simoncelli Howard Hughes Medical Institute, and New

Parameters: Marginal Statistics

Distribution of intensity values is captured with the first through fourth mo-ments of both the pixels and the lowpass coefficients at each pyramid scale.

Note: A number of authors have used marginal histograms:

Faugeras ’80 (pixels), Heeger & Bergen ’95 (wavelet), Zhu etal. ’96 (Gabor).

15 parameters

1/02

Page 14: Eero Simoncelli Howard Hughes Medical Institute, and New

Parameters: Spectral

Periodicity and globally oriented structure is best captured by frequency-domainmeasures (Francos, ’93).

Can be captured by autocorrelation measurements (included in most texturemodels).

In our model: central 7×7 region of the autocorrelation of each subband pro-vides a crude measure of spectral content within each subband.

125 parameters

1/02

Page 15: Eero Simoncelli Howard Hughes Medical Institute, and New

Parameters: Magnitude Correlation

Coefficient magnitudes are correlated both spatially and across bands. We cap-ture this with local autocorrelation and cross-correlation measurements.

472 parameters

1/02

Page 16: Eero Simoncelli Howard Hughes Medical Institute, and New

Parameters: Phase Correlation

Phases of complex responses at adjacent scales are aligned near image “fea-tures”.

We capture this using a novel measure of relative phase:

φ( f ,c) =c2 · f ∗

|c|,

where f is a fine-scale coefficient, c is a coarse-scale coefficient at the samelocation.

96 parameters

Total parameters: 708

1/02

Page 17: Eero Simoncelli Howard Hughes Medical Institute, and New

Phase Correlation Example

input

real/imag mag/phase real/imag mag/phase

coarse

realimag

phase mag x18

realimag

phase mag x18

fine

realimag

phase mag x18

realimag

phase mag x18

rphase

realimag

phase mag x20

realimag

phase mag x18

1/02

Page 18: Eero Simoncelli Howard Hughes Medical Institute, and New

Implementation

(high)

(low)

(mid)buildcomplexsteerablepyramid

imposeautoCorr

impose subbandstats & reconstruct(coarse-to-fine)

imposevariance

Gaussiannoise

+

imposeskew/kurt impose

pixelstatistics

synthetictexture

Each statistic, φk(~I), is imposed by gradient projection:

~I′ =~I +λk~∇φk(I), s.t. φk(~I) = mk,

where mk are the parameter values estimated from the example texture.

1/02

Page 19: Eero Simoncelli Howard Hughes Medical Institute, and New

Example Synthesis Sequence

Initial 1 4 64

We cannot prove convergence. But in practice, algorithm converges rapidly(typical: 50 iterations).

Run time: 256×256 image takes roughly 20 minutes (500 Mhz Pentium work-station, matlab code)

1/02

Page 20: Eero Simoncelli Howard Hughes Medical Institute, and New

Examples: Artificial

1/02

Page 21: Eero Simoncelli Howard Hughes Medical Institute, and New

Examples: Photographic, Quasi-periodic

1/02

Page 22: Eero Simoncelli Howard Hughes Medical Institute, and New

Examples: Photographic, Aperiodic

1/02

Page 23: Eero Simoncelli Howard Hughes Medical Institute, and New

Examples: Photographic, Structured

1/02

Page 24: Eero Simoncelli Howard Hughes Medical Institute, and New

Examples: Color

Color is incorporated by transforming to YIQ space, and including cross-bandmagnitude correlations in the parameterization.

1/02

Page 25: Eero Simoncelli Howard Hughes Medical Institute, and New

Examples: Non-textures?

1/02

Page 26: Eero Simoncelli Howard Hughes Medical Institute, and New

Necessity: Marginal Statistics

original with without

Needed for proper distribution of intensity values (at each scale).

1/02

Page 27: Eero Simoncelli Howard Hughes Medical Institute, and New

Necessity: Autocorrelation

original with without

Needed for capturing periodicity and global orientation.

1/02

Page 28: Eero Simoncelli Howard Hughes Medical Institute, and New

Necessity: Magnitude Correlation

original with without

Needed for capturing periodicity local structure.

1/02

Page 29: Eero Simoncelli Howard Hughes Medical Institute, and New

Necessity: Relative Phase

original with without

Needed for capturing details of local structure (edges vs. lines), and shading.

1/02

Page 30: Eero Simoncelli Howard Hughes Medical Institute, and New

Julesz Counter-Examples

Examples with identical 3rd-order pixel statistics

Left: Julesz ’78; Right: Yellott ’93

1/02

Page 31: Eero Simoncelli Howard Hughes Medical Institute, and New

Spatial Extrapolation

Modification: incorporate an additional projection operation in the synthesisloop, replacing central pixels by those of the original.

1/02

Page 32: Eero Simoncelli Howard Hughes Medical Institute, and New

Scale Extrapolation

Modification: incorporate an additional projection operation in the synthesisloop, replacing coarse-resolution coefficients by those of the original.

1/02

Page 33: Eero Simoncelli Howard Hughes Medical Institute, and New

Texture Mixtures

Modification: choose parameter vector that that is the average of those associ-ated with two example textures.

1/02

Page 34: Eero Simoncelli Howard Hughes Medical Institute, and New

Conclusions

• A framework for texture modeling, based on that originally proposed byJulesz

• New texture model:

– based on biologically-inspired statistical measurements

– includes methodology for testing

– provides heuristic methodology for refinement

– can be applied to a wide range of problems

Further information: http://www.cns.nyu.edu/∼lcv/texture

1/02

Page 35: Eero Simoncelli Howard Hughes Medical Institute, and New

To Do

• Adaptive front-end transformation (e.g., Zhu et al ’96, Manduchi & Portilla’99)

• Eliminate redundancy of parameterization

• Applications: compression, super-resolution, texture interpolation, texturepainting...

1/02