Compression and Pipeline Inspection Data

1
Compression and Pipeline Inspection Data Wei
Ching THAMSupervisor Dr Sandra I. Woolley
School of Electronic Electrical Engineering
2
Overview
 
 

• My PhD (Research)
• Pipeline Inspection Data
• Why Compress?
• Initial Investigations Discussion of Results
• Proposed compression method Diagnostically
  Lossless

 
 
 
 
 
3
My PhD (Research)
Investigate new and efficient methods for the
compression of pipeline data, so that
significantly more pipeline data can be stored on
each inspection tape.
4
Why Inspect Pipelines?
 
 
 
 
 
 
 
 
5
PIGs
 
 
Intelligent Pipeline Inspection Gauges PIGs
PIG size ranges from 12 to 56 with 2 increment
 
 
 
 
 
6
Pipes Cracks?
 
 
 
 
 
raw image
 
Signals from a seam weld, and a section through
the crack (62 x 100 mm long).
 
 
7
Magnetic Flux Leakage vs. Transverse Field
Inspection
 
 
False colour images showing a comparison of axial
and transverse field over the same regions of pipe
 
 
8
Why Compress?
 
TFI Systems
Average length of pipes 300km
1 Tape 12 GB (DDS3) DAT
 
 
On board electronics unit pre-processes data and
stores it for later retrieval
 
 
9
Compression

• Two fundamental types
• Lossless exact reproduction of original
• Traditionally preferred in data-sensitive
  applications
• but disappointing compression ratios
• Lossy approximation to original
• Loss of data might affect the diagnosis but good
  compression ratio

Diagnostically Lossless (a hybrid method
lossless and lossy) preserve all diagnostically
important image features
10
First Order Differencing
256
11
Lossless Compression Methods

• Huffman Coding takes advantages of the
  repetition of characters in a data streams
  (characters which occur often are allocated
  shorter codeword)
• Arithmetic Coding similar philosophy to Huffman
  but replaces a stream of input symbols with a
  single floating point output number. Slightly
  more efficient than Huffman but no tolerance to
  channel errors.
• Lempel-Ziv takes advantage of the repetition of
  strings in the data stream (DCLZ is a form of
  Lempel-Ziv 78)

12
Lossy Compression Methods

• DCT-based compression algorithm uses Discrete
  Cosine Transform
• Multi-Threshold Wavelet Codec uses Discrete
  Wavelet Transform

13
Discrete Cosine Transform
Transform image blocks into the frequency domain
Increased frequency from the top left to the
bottom right of the matrix.
14
DCT-based Compression
Linear Coefficient Quantization
Discrete Cosine Transform
Zig-zag Scanning
Run-Length Encoding Compression
15
Discrete Wavelet Transform
Transform data into different frequency components
16
Multi-Threshold Wavelet Codec

• enables control of 3 important parameters
• input image resolution (bits/pixel)
• output bit rate (bits/pixel) and
• the number of transform steps.

17
Image Quality Metrics

• Objective
• computational (MSE, SNR)
• Subjective
• performed by statistical analysis on viewers
  scores

Need to develop useful objective quality
metrics. i.e To define important features of
defects and develop methods which accurately
measures losses to these description.
18
Initial Results (1)
19
Initial Results (2)

• The best lossless
• vertical differencing followed by adaptive
  arithmetic coding giving 51 (1.63 bpp)
• The best lossy
• wavelet lossy_2 (MTWC with higher threshold)
  giving 161 (0.5 bpp)

20
Findings

• First-order differencing improves the
  compression performance significantly

• Adaptive arithmetic coding might be a good
  candidate for lossless coding

• Lossy compression methods may enable
  substantially improved compression performance
  with acceptable quality where acceptable quality
  is one that allows accurate diagnosis

21
Diagnostically Lossless

• Proposed diagnostically lossless scheme
• Identify pipe feature (possible defects)
• Bound the feature as shown (determining efficient
  and robust boundaries will be important)
• Compress bounded feature losslessly (or with
  minimal loss to feature description) and compress
  remaining data with high-quality wavelet method.

22
Why Wavelet? Not DCT (1)

• JPEG baseline uses DCT
• removal of high-frequency components (i.e.
  important pipeline defect information)
• designed for real-world human viewable images -
  not for instrumentation data used for computation
  of safety-critical features.

23
Why Wavelet? Not DCT (2)
24
Thank you! Questions?