A Novel Scheme for Video Similarity Detection

1
A Novel Scheme for Video Similarity Detection
Chu-Hong Hoi, Wei Wang  and Michael R. Lyu Dept.
of Computer Science Engineering, The Chinese
University of Hong Kong, Hong Kong, China

• Two main challenges for video similarity
  detection
• To find effective similarity measure metrics and
  techniques
• To detect similar videos from large video
  databases efficiently
• Our proposed solution
• We present a new two-phase similarity detection
  scheme based on two kinds of signatures with
  different granularities
• Coarse and Fine.
• Most of unrelated videos are filtered out with
  respect to the coarse similarity measure in the
  first phase.
• In the second phase, the query sample is
  compared with the results from the first phase
  using the fine similarity measure.

Filtering
Coarse Similarity Measure

• Coarse signature
• Pyramid Partitioning and Density Histogram (Fig
  1 (b))
• Naïve Pyramid Density Histogram (NPDH)
• Fuzzy Pyramid Density Histogram (FPDH)
• Coarse similarity Filtering
• Filtering out unrelated videos based on the PDH
  signatures

Fig 1. (a) Regular Partition (b) Pyramid
Partition
PDH vector
Fine Similarity Measure

• Fine signature
• Generating simplified video trajectories based
  on local similarity
• Fine similarity measure
• Dissimilarity measure between trajectories
• Finding the query results based on the Nearest
  Feature Trajectories

Conclusion The proposed two-phase scheme can
solve the video similarity detection effectively.