Folie 1

1
YUV Color System and SPIHT Coding
13th International Conference
Information Resources and Technologies
Topic
ANALYSIS OF THE YUV COLOR SYSTEM CONCERNING
OPTIMAL COMPRESSION OF COLOR IMAGES USING THE
SPIHT CODING ALGORITHM
Kathrin Sander (TU Ilmenau, Germany) -1-
2
YUV Color System and SPIHT Coding
Contents
1) SPIHT Process (Overview)
2) Better Compression Results using YUV
3) Need of a common Bitrate Ratio
4) Finding Static Format for YUV
5) Proving Static Format by Subjective Tests
6) PSNR subjective Image Quality Perception
Kathrin Sander (TU Ilmenau, Germany) -2-
3
1) SPIHT Process (Overview) 1
SPIHT Process
Kathrin Sander (TU Ilmenau, Germany) -3-
4
1) SPIHT Process (Overview) 2
Ways to improve Compression Results
Conventional

• Better Wavelet Filters
• Better Coding Methods

New

• Modified Input Data being more capable
• for the Coding Process (RGB -gt YUV)

Kathrin Sander (TU Ilmenau, Germany) -4-
5
2) Better Compression Results using YUV 1
RGB to YUV
YUV as Standard for Color TV Transmission
Y (Brightness Perception of human Eye)
U (weighted Color Difference B-Y)
V (weighted Color Difference R-Y)
Kathrin Sander (TU Ilmenau, Germany) -5-
6
2) Better Compression Results using YUV 2
Histogram Example hut.bmp
RGB
R
G
B
YUV
Y
U
V
Kathrin Sander (TU Ilmenau, Germany) -6-
7
2) Better Compression Results using YUV 3
Advantages YUV

• U and V allocate only between ¼ and ½ of the
• available Value Domain (28256)

• Less bits neccessary to code Detail Infor-
• mation at same Quality as when using RBG

• Worldwide common digital Video Standards
• Compression Algorithms (i.e. MPEG2) based
• on YUV

Kathrin Sander (TU Ilmenau, Germany) -7-
8
2) Better Compression Results using YUV 4
Value Distribution YUV 1
Aim

• Split available total bitrate into 3 parts
• of various size depending on the value
• distributions in the single Image Channels

Kathrin Sander (TU Ilmenau, Germany) -8-
9
2) Better Compression Results using YUV 5
Value Distribution YUV 2

• Former Research Work at MPEI Energy
• describes non Gaussian value distribution
• in Y, U V channels

• Real Bitrate Ratio may be calculated
• based on Energy

Kathrin Sander (TU Ilmenau, Germany) -9-
10
2) Better Compression Results using YUV 6
Value Distribution YUV 3

• Value distributions in Y, U V are similar
• (!not the same!) for all images -gt calculated
• Bitrate Ratios are different for all Images

Kathrin Sander (TU Ilmenau, Germany) -10-
11
3) Need of a common Bitrate Ratio 1
Disadvantages Real Format (RF)

• calculating RF before coding each image
• -gt very time consuming

• Tests RF just Orientation, not always best
• Compression Results (as when using Optimal
• Format (OF))

Kathrin Sander (TU Ilmenau, Germany) -11-
12
3) Need of a common Bitrate Ratio 2
Static Format instead of Real Format
Aim

• finding a general Static Format (SF)
• that gives good Results independent from
  Parameters like Image Content, Bitrate, Image
  Size and used Wavelet Filter

Kathrin Sander (TU Ilmenau, Germany) -12-
13
4) Finding Static Format for YUV 1
Determining Static Format 1

• Finding SF by Setting up an Equation
• impossible (too many Parameters)

• Instead Finding SF empirically by
• Running Tests (using MATLAB EXCEL)

• Compression Quality Criterion PSNR

Kathrin Sander (TU Ilmenau, Germany) -13-
14
4) Finding Static Format for YUV 2
Determining Static Format 2
Test Parameters

• 11 Images with different
• Contents at 2 Sizes
• (256x256 512x512)

• 4 Bitrates (0.1, 0.4, 0.7, 1.0)

• 3 Wavelet Filters

• 1,683 potential Formats (111 till 201010
• in Steps of 1, leaving out repeated ones)

Kathrin Sander (TU Ilmenau, Germany) -14-
15
4) Finding Static Format for YUV 3
Determining Static Format 3
Test Results

• 4x3x11x2264 Lists with 1,683 Formats sorted
  decreasing by their appropriate PSNR were
  evaluated statistically

• Not one SF -gt formats about 75.012.512.5

Kathrin Sander (TU Ilmenau, Germany) -15-
16
4) Finding Static Format for YUV 4
Determining Static Format 4
Test Results for SF 73.715.810.5

• for 74.2 of Result Lists
• PSNR(SF) gt PSNR(RF)

• Averaged PSNR using SF just 1.03
• (0.26db) less than with respective OFs

• max PSNR difference using SF OF 1.9db

• min PSNR difference using SF OF 0.0db

Kathrin Sander (TU Ilmenau, Germany) -16-
17
5) Proving Static Format by Subjective Tests 1
Need of Subjective Tests

• PSNR objective Criterion, that does not
• necessarily reflect human Perception of
• Compression Quality

• Subjective Tests to prove the Practicality
• of found SF

Kathrin Sander (TU Ilmenau, Germany) -17-
18
5) Proving Static Format by Subjective Tests 2
Test Sequence 1
Norm

• Based on ITU-R BT.500-11 Norm
• about Methodology for the subjective
• Assessment of the Quality of TV Pictures

Kathrin Sander (TU Ilmenau, Germany) -18-
19
5) Proving Static Format by Subjective Tests 3
Test Sequence 2
Test Parameters

• 32 Persons

• 11 Pictures (512x512)

• 16 Quality Stages
• (sorted worst till best)

• 12 different Bitrates
• (12x SF, 4x respective OF)

• Filter 2bior44 (best PSNR)

Kathrin Sander (TU Ilmenau, Germany) -19-
20
5) Proving Static Format by Subjective Tests 4
Test Sequence 3
Users Task

• Quality Rating of all Reconstruc-

tions using a Scale from 1 (excellent) to 5
(bad)
Kathrin Sander (TU Ilmenau, Germany) -20-
21
5) Proving Static Format by Subjective Tests 5
Test Results

• 81 of Compressed Pictures using SF and
• respective OF at the same Bitrate
• -gt perceived to have same Quality

• Perceived Quality Difference
• never gt one Level of Scale

• sufficient good Compression Results for SF

Kathrin Sander (TU Ilmenau, Germany) -21-
22
6) PSNR subjective Image Quality Perception 1
Does Coherence between PSNR subjective Image
Quality Perception exist?

• Answer gained by comparing described
• objective subjective Tests

Kathrin Sander (TU Ilmenau, Germany) -22-
23
6) PSNR subjective Image Quality Perception 2
Test Results 1
Most Images

• bad Quality lt 25db

• excellent Quality gt 32db

• in between
• linear
• Increase

Kathrin Sander (TU Ilmenau, Germany) -23-
24
6) PSNR subjective Image Quality Perception 3
Test Results 2

• better subjective Quality at
• lower or higher PSNR depending

Exceptions
on different levels of detail within Images
Kathrin Sander (TU Ilmenau, Germany) -24-
25
YUV Color System and SPIHT Coding
D A N K E S C H Ö N
)
THANK YOU
??????? ???????
Kathrin Sander (TU Ilmenau, Germany) -25-