Texture Loss for JPEG 2000 Compression

Measurement of Texture Loss for JPEG 2000 Compression 

Peter D. Burns and Don Williams*Burns Digital Imaging and *Image Science Associates

Presented at IS&T and SPIE Electronic Imaging Symposium, Jan. 2012© copyright 2012 Peter D. Burns

Full paper available here

IS&T and SPIE Electronic Imaging 2012 2

Introduction

MTF established as a metric for the capture and retention of image detail

Texture-loss MTF using targets with random objects• Dead-leaves target analysis based on noise-power

spectrum

We apply this method to image detail loss during image compression

Adapt method when printed test target is not used

Compare results for JPEG2000 and JPEG with Structured Similarity Index (SSIM)

IS&T and SPIE Electronic Imaging 2012 3

Dead-Leaves MTF Measurement

Aimed at providing an effective MTF for image fluctuations (signals) influenced by adaptive or signal-dependent image processing

• e.g., adaptive noise cleaning, which could leave edge untouched, but reduce detail in important ‘textured regions’

Being developed as part of the CPIQ Initiative

Based on input and output Noise-power spectrum

filterednoisy

IS&T and SPIE Electronic Imaging 2012 4

Texture MTF using Noise-power Spectrum*

Printed Test chart

Digital image

One-dimensional Noise-power spectra

0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5

102

103

104

Frquency, cy/mm

Pow

er S

pect

rum

Input target

JPEG 2000

Digital camera, image processing

____________________* Also called power spectral density

0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.50

0.2

0.4

0.6

0.8

1

Frquency, cy/mm

MT

F txt

Texture MTF

IS&T and SPIE Electronic Imaging 2012 5

Noise-power Spectrum: meaning and measurement

• Noise-Power spectrum: for a random process, the NPS describes the fluctuations as a function of spatial frequency

Technically: Fourier transform of the spatial autocovariance

• Measurement: Average square of the Discrete Fourier Transform of a nominally uniform data array

Select data array

Compute2D FFT

Compute modulus squared

Basic steps for NPS estimation

1 or 2D

Vari

ance

/fre

quency

frequencyo

FineCoarse

IS&T and SPIE Electronic Imaging 2012 6

Recipe:

Transform the captured image data to luminance

Compute the power-spectral density as the square of the amplitude of the two-dimensional DFT of the array

Divide this array, frequency-by-frequency, by the spectrum for the input target

Compute the square-root, frequency-by-frequency

Radial-average of this array is the one-dimensional MTF vector

Compute (visually -weighted) acutance measure

Proposed Dead-Leaves MTF Measurement

IS&T and SPIE Electronic Imaging 2012 7

Proposed method for camera evaluation (basic steps)

Printed targetDigital image

Transform to luminance

signal spectrum

1. Compute corrected signal spectrum

2. Texture MTF

0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.50

0.2

0.4

0.6

0.8

1

Frquency, cy/mm

MT

F txt

Model target signal spectrum, Starget

3. Texture acutance metric

Acutancemetric

texture MTF

IS&T and SPIE Electronic Imaging 2012 8

Modified method (details)Input image

Output imageTransform to

luminance

2. Texture MTF

0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.50

0.2

0.4

0.6

0.8

1

Frquency, cy/mm

MT

F txt

Measured input signal spectrum, Starget

targetsignal SSR /'

vR

vMTFrad

3. Texture acutance metric

vCSFvMvMTFv

vrad

max

1

Acutancemetric

Visually-weighted summation*___________* Display MTF and viewing distance

Corrected signal spectrum

1. Compute corrected signal spectrum

2D FFT 2D FFT

Noisespectrum

noisesignal SS

Signalspectrum

Radial integration0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5

102

103

104

Frquency, cy/mm

Pow

er S

pect

rum

Input target

JPEG 2000

IS&T and SPIE Electronic Imaging 2012 9

Application to Image Compression

Detectornoise

CFAsubsampling3 1 31

CFAinterpolation

Optical MTF

Input idealimage

Simulated captured image

Detectornoise

CFAsubsampling33 11 3311

CFAinterpolation

CFAinterpolation

Optical MTFOptical MTF

Input idealimage

Simulated captured image

ImageCompression

Image capture simulation

IS&T and SPIE Electronic Imaging 2012 10

JPEG 2000 and JPEG compression

JPEG2000: kdu_compress, from Kakadu Software

JPEG: as implemented in Matlab

Default settings for 24-bit color images

Compression ratios: up to 140:1

input 40:1 100:1

Example:

IS&T and SPIE Electronic Imaging 2012 11

Example texture MTF

0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5

102

103

104

Frquency, cy/mm

Pow

er S

pect

rum

Input target

JPEG 2000

0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.50

0.2

0.4

0.6

0.8

1

Frquency, cy/mm

MT

F txt

Results for 100:1 compression, and the corresponding texture MTF.

acutance = 0.82

IS&T and SPIE Electronic Imaging 2012 12

Comparison with Structured Similarity Index, SSIM

JPEG 2000 JPEG

Compression rate

Bits/pixel/ color

Texture acutance

SSI index Texture acutance

SSIM index

30 0.80 0.986 0.939 1.02 0.939

40 0.60 0.950 0.924 0.991 0.922

50 0.48 0.951 0.920 0.961 0.904

60 0.40 0.859 0.908 0.930 0.886

80 0.30 0.826 0.884 0.890 0.844

100 0.24 0.819 0.865 0.841 0.790

120 0.2 0.796 0.835 0.777 0.742

140 0.17 0.731 0.799 0.667 0.667

0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 10.6

0.65

0.7

0.75

0.8

0.85

0.9

0.95

1

1.05

SIMM value

Tex

ture

acu

tanc

e

JPEG 2000

JPEG• objective measure of image quality• based on image differences • visual-difference map, based on a

model of visually information• average value of the difference image

is reported as the SSIM. Wang, Z., Bovik, A.., Sheikh, H., and Simonelli, E., IEEE Trans. Image Processing, (2004)

IS&T and SPIE Electronic Imaging 2012 13

Summary

Many practical objective image quality measurements can be considered as estimates, with bias error and variation

The proposed texture MTF analysis relies on noise-power spectrum estimation

We investigated texture-loss due to JPEG 2000 and JPEG compression

Modified method was developed;• Direct input signal spectrum measurement (estimation)• Not dependent on known printed target spectrum

Results indicated stable texture MTF and acutance without date smoothing or fitting

Compared well with Structured Similarity Index, SSIM• an offset between the JPEG and JPEG 2000 images sets

[email protected]

IS&T and SPIE Electronic Imaging 2012 14

Appendix: Example MTF based on Edge SFR and texture NPS

Edge SFR

Comparison with Texture MTF: Results for 100:1 compression ratio

0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.50

0.2

0.4

0.6

0.8

1

Frquency, cy/mm

MT

F txt

- - - Edge

___ Texture

0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.50

0.2

0.4

0.6

0.8

1

Spatial frequency, cy/pixel

SF

R

input

output

0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.50

0.2

0.4

0.6

0.8

1

Spatial frequency, cy/pixel

MT

F