Design and Selection of Access Points

• Most systems provide content features that are computable in reasonable cost
  image color, texture, object shape, position, range, video key frame, scene change, ...
• Different applications need different access points
  • Improper access points reduces precision
  • Excessive access points leads to system overload
• Ramesh Jain classify content features into 3 categories:
  • metadata
    • size, resolution, author, date, ...
    • those that can not derived from data content
  • derived features
    • precomputed features before storage and indexing
    • require special algorithms or semi-automatic processing
  • computed features
    • applied after the search space has been narrow down
    • request on demand

前一張

下一張

回到第一張

瀏覽圖案部份