University of Freiburg ETeam Visual Saliency

1
University of Freiburg E-Team Visual Saliency
Gerd Brunner, Alexandra Teynor, Lokesh Setia

• Expertise of UFR
• Content Based Image Retrieval (CBIR)
• Relevance Feedback Methods (SVM, weighting)
• Weak segmentation for CBIR
• Recognition of specific objects and visual
  object classes
• Segmentation of 3D data (different cell types)
• Pollen classification (2D/3D)
• Projects
• I-Search , Simba, MICHELscope, OMNIBUSS, OMIB

Simba
MICHELscope
2

• How we do it
• Focus on features
• Local invariant features (using Haar integrals)
• Features evaluated at salient points or regions
  (patches)
• Wavelet based features (JPEG 2000)
• Invariant line segment features

3

• Current Research
• Visual object class recognition
• Evaluation of properties of patch based
  approaches
• Image classification (building non building)
• CBIR for Watermarks (Filigrana)
• Next focus
• Partial image retrieval/matching
• Incorporation of spatial arrangements
• of features (patches, line segments )
• Automatic image annotation