Compression for Synthetic Aperture Sonar Signals

1
Compression for Synthetic Aperture Sonar Signals

• Thomas Higdon
• MDDSP
• Feb. 25, 2008

2
What is synthetic aperture processing?

• Collect sensor data at a series of physical
  locations.
• Aggregate the data and process it to form an
  image.

3
Typical SAS Image

• Edge detection
• Speckle noise reduction

4
Why compression is needed

• Data for a typical sonar array might arrive at
  many gigabytes/sec.
• Storage on autonomous vehicles is limited.
• Compression might allow data to be reasonably
  transmitted via underwater communication links.

5
Radar and ultrasound techniques

• Applied to radar and ultrasound images
• Sonar shares noise characteristics with radar and
  ultrasound.
• Application of wavelet-based techniques proposed
  for ultrasound and radar to sonar.

6
SPIHT
Said, Pearlman,. 2005

• Wavelet transform-based
• Transmits wavelet coefficients with more
  information first.
• Capable of very low bit rates by recording only
  decisions made by the encoder.
• Capable of arbitrary bit rates.

7
SPIHT at 161 compression
8
Modified SPIHT for SAR
Zeng, Cumming, 2001

• Uses wavelet packet transform
• Prunes wavelet packet tree for a given threshold.
• Also encodes texture information more
  efficiently.
• Uses soft-thesholding to reduce speckle noise.

9
Ultrasound Technique
Gupta, et al. 2005

• Logarithm to convert multiplicative speckle noise
  to additive noise.
• Perform wavelet transform and estimate quantizer
  thresholds and subdivide coefficients into
  classes based on activity level.
• Threshold and quantize the coefficients in each
  class using an adaptive uniform threshold
  quantizer.

10
Results 401 compression
11
Image Processing

• The reduction of noise that does not contain
  image information will allow more efficient
  compression.
• Conversion to log space to allow reduction of
  multiplicative noise.

12
Proposed Method

• Leverage SPIHT algorithm
• Modify it to suit SAS sonar images.
• Use wavelet packet transform to determine
  information about edges and store it efficiently.
• Speckle noise.

13
Image Assessment

• Evaluation of the performance of different
  techniques requires a metric.
• Sonar images typically processed by human
  operators and/or object-detection algorithms.
• Multiple image assessment techniques will be used
  to determine the difference between the
  compressed and uncompressed images.

14
Questions