Visual Evaluation of JPEG 2000 Color Image Compression Performance N1583

1
Visual Evaluation of JPEG 2000 Color Image
Compression Performance (N1583)

• Testing Group
• Troy Chinen, Fujifilm Software California
• Alan Chien, Eastman Kodak

2
Testing Goals

• Quantitative
• For a given set of JPEG 2000 bitrates, compute
  the corresponding quality-equivalent JPEG
  bitrates for color images.
  
• Qualitative
• Characterize JPEG 2000 color compression
  artifacts.

3
Testing method
Ordered JPEG images
1
2
3
4
5
6
7
8
9
10
0.3bpp 0.4 0.5 0.6 0.7
0.8 0.9 1.0 1.2
1.4
C(0.25bpp)
B(0.5bpp)
A(0.75bpp)
D(1.0bpp)
VL
Unordered JPEG 2000 images

• Insert JPEG 2000 images between JPEG images of
  bounding quality.
  
• Record mean number of bytes for each insertion
  fit to Gaussian distribution

4
Testing Method (ctd)

• 6 scenes bike, bottles, bride, harbor, woman,
  wool
  
• 10 Reference JPEG images per scene
• default mode cjpeg, optimized Huffman tables
• 0.3 1.0 (0.1 increments), 1.2, 1.4bpp
• 4 JPEG 2000 images per scene
• Default mode, VM5.2
• Fweights table T15PB
• 0.25, 0.50, 0.75, 1.0 bpp

5
Testing Method (ctd)

• Hardcopy 300 dpi, 24 bit color
• Blind testing
• 6 Observers
• VL (visually lossless) rankings were not included
  in the summarizing statistics

6
Results aggregate
11
18
36
53
Error bars 2 sigma
7
Results bike
8
Results bottles
9
Results bride
10
Results harbor
11
Results woman
12
Results wool
13
Qualitative Results

• No color distortions observed
• Most JP2 images at 1.0bpp were VL (in print)
• Trends

14
Conclusions

• JPEG 2000 filesizes are smaller than JPEG by
• 53 at 0.25bpp
• 36 at 0.5 bpp
• 18 at 0.75 bpp
• 11 at 1.0 bpp
• Distortion types of JPEG and JPEG 2000 are
  fundamentally different, nevertheless consistent
  rankings were observed.