Seafloor Image Compression and Large Tilesize Vector Quantization - - PowerPoint PPT Presentation

▶
seafloor image compression and large tilesize vector
SMART_READER_LITE
LIVE PREVIEW

Seafloor Image Compression and Large Tilesize Vector Quantization - - PowerPoint PPT Presentation

Aug 14, 2022 333 likes •762 views

Seafloor Image Compression and Large Tilesize Vector Quantization Chris Murphy Robert Y. Wang Dr. Hanumant Singh AUV Conference, 2 September 2010 Environmental Sensors 2 AUV Conference, 2 September 2010 Sidescan Sonar Imagery 3 AUV

slide-1
SLIDE 1

AUV Conference, 2 September 2010

Seafloor Image Compression and Large Tilesize Vector Quantization

Chris Murphy Robert Y. Wang

  • Dr. Hanumant Singh
slide-2
SLIDE 2

AUV Conference, 2 September 2010

2

Environmental Sensors

slide-3
SLIDE 3

AUV Conference, 2 September 2010

3

Sidescan Sonar Imagery

slide-4
SLIDE 4

AUV Conference, 2 September 2010

4

Photographs

slide-5
SLIDE 5

AUV Conference, 2 September 2010

5

The Problem

Data is typically unavailable until after recovery. Effective collaboration requires information sharing, whether from sub-sea to surface or sub-sea to sub-sea. In real-world conditions, effective throughput of modern acoustic modems can be 10-100 bits per second. How can we effectively share multimodal information at these rates?

slide-6
SLIDE 6

AUV Conference, 2 September 2010

6

Two Complimentary Approaches

Large Tilesize Vector Quantization SPIHT

slide-7
SLIDE 7

AUV Conference, 2 September 2010

7

Concept: LTVQ

Frequently, we have imagery available from previous dives that is similar in content; possibly from the same geographic location. How can these previous photos be used to Improve image compression?

slide-8
SLIDE 8

AUV Conference, 2 September 2010

8

Vector Quantization

Basic Idea:

  • Encode samples by referencing

indices of previously established “codewords”. The appropriate codeword is selected as the “closest” using some metric, such as L2 (Euclidean) distance.

  • 1

2 3 4 5 1, 2 4, 1.8 1.2, 2.1

slide-9
SLIDE 9

AUV Conference, 2 September 2010

9

Vector Quantization

Basic Idea:

  • Encode a large number of samples

simply by referencing the index of a previously established “codeword” which is closest using some metric, such as L2 distance.

  • Early image compression algorithms

used this technique to encode small (eg: 4x4) squares of pixels.

  • 1

2 3 4 5 1,2 4, 1.8

slide-10
SLIDE 10

AUV Conference, 2 September 2010

10

Compressed Image Size

Compressing 4x4 blocks of pixels can only go so far. How can we use VQ to obtain very high compression ratios?

slide-11
SLIDE 11

AUV Conference, 2 September 2010

11

Inspiration

slide-12
SLIDE 12

AUV Conference, 2 September 2010

12

Compressed Image Size

So – how can we use VQ to obtain very high compression ratios? Simple - increase the size of the tiles!

slide-13
SLIDE 13

AUV Conference, 2 September 2010

13

Large Tilesize Vector Quantization

Even relatively large seafloor image tiles (here, 24 pixels square) exhibit homogeneity. Previous dive imagery can be sliced into tiles to obtain codewords for future dives.

slide-14
SLIDE 14

AUV Conference, 2 September 2010

14

Example

All tiles taken from 150 randomly selected images, collected on a previous dive.

slide-15
SLIDE 15

AUV Conference, 2 September 2010

15

slide-16
SLIDE 16

AUV Conference, 2 September 2010

16

Codeword Matching

Problem: The number of possible codewords (or “tiles”) generated from even a single dive can be huge. Pixel-by-pixel comparison against each of them is computationally impractical, especially for an AUV. 24 x 24 x 3 = 1728 comparisons per tile.

slide-17
SLIDE 17

AUV Conference, 2 September 2010

17

Codeword Matching

Solution: Calculate principle components for the tiles using

  • PCA. Match principle component weights to reduce

dimensionality, rather than tiles directly.

slide-18
SLIDE 18

AUV Conference, 2 September 2010

18

De-Blocking

Reconstructed images exhibit heavy blocking artifacts at tile boundaries. How can those artifacts be reduced

  • r eliminated?
slide-19
SLIDE 19

AUV Conference, 2 September 2010

19

“Image Quilting” / MinCuts

Minimize error at tile borders by carefully selecting tile boundary. Efros + Freeman, “Image Quilting for Texture Synthesis and Transfer,” SIGGRAPH 2001.

slide-20
SLIDE 20

AUV Conference, 2 September 2010

20

Poisson Image Editing

Interpolate within boundaries, while preserving gradients. Perez et al., “Poisson image editing,” ACM Trans. Graph., 2003.

slide-21
SLIDE 21

AUV Conference, 2 September 2010

21

slide-22
SLIDE 22

AUV Conference, 2 September 2010

22

slide-23
SLIDE 23

AUV Conference, 2 September 2010

23

SPIHT

Set Partitioning In Hierarchical Trees *

Works in Wavelet Domain

  • Wavelets have been shown to provide an efficient, sparse representation

for a variety of real-world signals. Dimensionality-Independent

  • Same encoding works for CTD data, Imagery, Volumetric data

Embedded Coding

  • Low-fidelity versions are identical to the beginning of high-fidelity
  • Each additional (in-order) byte improves the estimate
  • Sending up a higher quality version doesn't require 'starting over'

* Said and Pearlman, 1996

slide-24
SLIDE 24

AUV Conference, 2 September 2010

24

Fully Embedded Coding

Fully embedded coding – A high quality version of encoded data shares first N bits identically with a poor quality encoding of length N. Low Quality Preview Medium Quality Representation High Quality Representation If preview is interesting, recipients can request additional packets to obtain high quality sections of signals or images.

slide-25
SLIDE 25

AUV Conference, 2 September 2010

25

Reduction Potential - 28 Bytes

slide-26
SLIDE 26

AUV Conference, 2 September 2010

26

Reduction Potential - 56 Bytes

slide-27
SLIDE 27

AUV Conference, 2 September 2010

27

Reduction Potential - 112 Bytes

slide-28
SLIDE 28

AUV Conference, 2 September 2010

28

Temperature - 28 Bytes

slide-29
SLIDE 29

AUV Conference, 2 September 2010

29

Temperature - 56 Bytes

slide-30
SLIDE 30

AUV Conference, 2 September 2010

30

Temperature - 112 Bytes

slide-31
SLIDE 31

AUV Conference, 2 September 2010

31

Sidescan Imagery

1008 Bytes 4032 Bytes

slide-32
SLIDE 32

AUV Conference, 2 September 2010

32

Sidescan Imagery

1008 Bytes 4032 Bytes

slide-33
SLIDE 33

AUV Conference, 2 September 2010

33

Sidescan Imagery

315 Bytes 1008 Bytes

slide-34
SLIDE 34

AUV Conference, 2 September 2010

34

Live Trials near Rota

North of Guam, 1600mi east of the Philippines

slide-35
SLIDE 35

AUV Conference, 2 September 2010

35

Original

slide-36
SLIDE 36

AUV Conference, 2 September 2010

36

4032 Bytes

slide-37
SLIDE 37

AUV Conference, 2 September 2010

37

slide-38
SLIDE 38

AUV Conference, 2 September 2010

38

slide-39
SLIDE 39

AUV Conference, 2 September 2010

39

6048 Bytes 7056 Bytes 8064 Bytes 5040 Bytes 10080 Bytes 11088 Bytes 12096 Bytes 9072 Bytes 2016 Bytes 3024 Bytes 4032 Bytes 1008 Bytes

slide-40
SLIDE 40

AUV Conference, 2 September 2010

40

A Side Note About Compression

It is easier to compress more (correlated) data than less. Thus, the less often you share information the more efficiently you can share it. How does this affect our communication strategy?

slide-41
SLIDE 41

AUV Conference, 2 September 2010

41

Thank You.

chrismurf @ whoi.edu

slide-42
SLIDE 42

AUV Conference, 2 September 2010

42

Brief Introduction to SPIHT

Sorting Bits Indicate

  • Is a coefficient “significant”?
  • Are any descendants?
  • Any grand-descendant?

Refinement Bits Indicate

  • The sign of a coefficient
  • Coefficient magnitude