outline

1
outline

• Two region based shape analysis approach
• Moment and moment invariants
• Wavelet based method combined with moment based
  method
• Combination of various shape descriptors
• Future work

2
Region based shape analysis

• Graphical objects are represented by a planar
  graph with nodes representing sub-regions.
• region skeleton
• region decomposition
• Scalar computer scalar result based on
  global shape, global transform descriptors
  include moments,Fourier, Walsh etc.
• moment based methodmost popular
• shape Matrices and vector
• mathematical morphologysuitable for
  shape related processing.
• Shortcomings of global scalar transform
• can not measure the degree of similarity
• can not match query with part of image
• sensitive to noise and occlusion
  

3
Moment based method

• Moment
• Moment is used to calculate
  statistical data of geometric properties of
  distribution such as area, centroid
  ,moment of inertia, skewness,
• Mathematical presentation
• moment m of order of pq of
  function f(x,y) is
• Advantage of moment based method
• information preserving---moment m is uniquely
  determined by f(x,y), vice versa, m can be used
  accurately reconstruct f(x,y) .
• mathematically concise.
• Disadvantage of moment based method
• difficult to correlate high
  order moments with shape feature.

4
Moment invariants

• Moment invariants
• fundamental moment formula is not invariant to
  translation,rotation and scaledepending on
  position,orientation,or scale.
• Hus 7 normalized central moment invariants is
  the foundation for latter application in 1961.
• Orthogonal moments(Legendre,Zernike,etc.) is
  superior to regular moments, complex moments in
  terms information redundancy . Zernike moments
  have the the best overall performance.
• Fuzzy moment in order to separate object and
  background into different class ,apply fuzzy
  logic to obtain optimal parameter.
• DOH to DOM
• DOH(difference of histogram) is suitable
  for real time (not sensitive to motion), but is
  sensitive to translation and scale.
• DOM(difference of moments) moments invariants
  are giving good performance when lighting
  condition changes.

5
Wavelet based method combined with moment method

• Wavelet based method wavelet transform can
  provide multiresolution capability and high
  compaction.
• wavelet based compressor and
  decompressor

Compressor
Forward wavelet transform
Original image
Quantizer
Encoder
Decompressor
Decompressed image
Inverse Wavelet transform
Dequantizer
Decoder
6
wavelet based illumination invariant indexing

• TSI-LGMWP
• M.K.mandal proposed TSI-LGMWPcombine moment
  technique(TSI-LGM) with wavelet technique(WP)
• Indexing is performed directly on compressed
  data, moment is used to improve compression
  efficiency.
• Image retrieval result

7
Combination of various descriptors

• Fourier and moment descriptors
• J.S.Park and colleagues use two
  stage scheme,
• 1. compute moments, 2. improved
  by Fourier descriptors,
• best result
• Hus moment invariants Fourier
• Zernike moment invariants Fourier
  descriptor
• Simple combined descriptors
• Jukka and colleagues compared CCH,
  PGH and combined simple descriptors of convexity,
  principle axes, compactness, variance and
  elliptic variance .
• result
• combined descriptors has medium
  performance on time and memory compared with the
  other two, but gives best recognition result when
  using small irregular objects as test .
  

8
Combination of Five simple descriptors
9
Future work

• Combined descriptors
• Human perceptual system compute
  similarity involves both region and boundary
  aspects.
• Segmentation extract only objects of interest
• Shape matching for partially recovered or objects
  having occlusion.
• Semantically meaningful retrieval develop
  perceptually based image features.