Lossy Compression of Color Mosaic Images

1
Lossy Compression of Color Mosaic Images

• EE398A Final Project Presentation
• 13th March, 2007
• Stephanie Kwan
• Karen Zhu

2
Mosaic Images
Bayer Pattern

• Digital cameras capture images using a color
  filter array (CFA)
• Only 1 color is captured in each sensor
• Most commonly used CFA pattern - Bayer Pattern

3
Example of Mosaic Image
4
Processing Mosaic Images

• Have to estimate the missing 2 color values in
  each pixel
• Demosaic - Interpolate a complete color image
  from the raw data
• Conventional Lossy Compression of Images

Demosaic
Color Balance
RGB image
Raw mosaic image
transmission, storage
Lossy Compression
Decompression
Compressed RGB image
5
Motivation

• Demosaicking- first and compression-later design
  increases algorithm complexity.
• New approach Compression-first and
  demosaicking-later design
• Specifically interested in performance under
  Lossy Compression

Lossy Compression
Decompression
Raw mosaic image
Compressed mosaic image
Demosaic
Color Balance
Compressed RGB image
6
Approach 1 Directly compress mosaic image
Wavelet Transform (5/3 wavelet)
SPHITE Arithmetic Encoding
Raw mosaic image
Lossy Compression
Inverse Wavelet Transform (5/3 wavelet)
SPHITE Arithmetic Decoding
Demosaic Color Balancing
Decompression
Final Compressed RGB image
7
Approach 2Compress R,G, B seprately
Wavelet Transform (5/3 wavelet)

Separate into 3 subimages R, G, B
Raw mosaic image
Lossy Compression
SPHITE Arithmetic Encoding
Combine subimages back to mosaic pattern

Demosaic Color Balancing
SPHITE Arithmetic Decoding
Inverse Wavelet Transform (5/3 wavelet)

Decompression
Final Compressed RGB image
8
Mallet Packet Decomposition
5 Level Mallet Packet Decomposition
Raw Mosaic Image
9
Results CPSNR PSNR Comparison
Reference Demosaiced, uncompressed RGB images
Reference Uncompressed Raw Mosaic Images
10
Results Visual Comparison
Result of Demosaicking Uncompressed Mosaic Image
Original Mosaic Image
11
Results Visual Comparison
Reference YUV Compressed Distortion of YUV
211 Compression ratio 14.99 CPSNR 32.63
Proposed Method (RGB combined) Compression ratio
14.45 CPSNR 30.074
12
Results Visual Comparison
Proposed Method (R,G,B separated) Distortion
ratio of RGB 111 Compression ratio 14.3
CPSNR 28.897
Reference RGB Compressed Distortion of RGB
111 Compression ratio 14.39 CPSNR 29.33
13
Results Visual Comparison
Approach 1(RGB combined) Compression ratio
12.29 CPSNR 31.125
Approach 2(RGB separated) Compression ratio
12.5 CPSNR 29.057
14
Conclusion

• When Lossy Compression is performed, under the
  same compression ratio, compression-first
  demosaic-later design produces better quality
  color images
• For our proposed approaches, directly compressing
  the mosaic image, or separately compressing R,G,B
  subimages gives similar performance. Which
  approach is better depends on personal preference
  and the image used
• Future works Explore results of transforming
  R,G,B subimages using DPCM instead of wavelet
  transform

15
Thank You!
16
Appendix