Lossy Image Compression Results

Lossy compression results on the New Test Images.

Results measure visual-quality of images compressed at different bit-rates by various algorithms. Higher SSIM value means better visual quality of compressed image. Higher quality at same bit-rate means better results. Click on the images to enlarge, you can enlarge multiple images at same time. You can also download numbers as tab separated text file.

artificial.ppm

3072x2048

Compression - Visual Quality

big_building.ppm

7216x5412

Compression - Visual Quality

big_tree.ppm

6088x4550

Compression - Visual Quality

bridge.ppm

2749x4049

Compression - Visual Quality

cathedral.ppm

2000x3008

Compression - Visual Quality

deer.ppm

4043x2641

Compression - Visual Quality

fireworks.ppm

3136x2352

Compression - Visual Quality

flower_foveon.ppm

2268x1512

Compression - Visual Quality

hdr.ppm

3072x2048

Compression - Visual Quality

leaves_iso_200.ppm

3008x2000

Compression - Visual Quality

leaves_iso_1600.ppm

3008x2000

Compression - Visual Quality

nightshot_iso_100.ppm

3136x2352

Compression - Visual Quality

nightshot_iso_1600.ppm

3136x2352

Compression - Visual Quality

spider_web.ppm

4256x2848

Compression - Visual Quality

zone_plate.ppm

3000x2000

Compression - Visual Quality

Implementations Used

• For Jpeg, IJG's open-source implenemtation v6b was used with libpnm executables (Parameters: -baseline -quality X)
• For Jpeg 2000, Kakadu v6.0 demo executables were used (Parameters: -rate X)
• For HD Photo, MS DPK v1.0 was used (Parameters: -c 9 -q X)
• For visual quality measurement, TU Berlin implementation of SSIM by Felix Oum and Thomas Richter was used

Let me know if you want to see some other algorithm/implementation included in the benchmark. Goal is not to be exhaustive and cover all algorithms/implementations, the results must be interesting enough.