Training

5-6 segmentations by different human subjects per image. Human-assisted selection of symmetric segments. Combination of skeletons for selected parts gives the final image ground-truth. Learning-based symmetry detection in natural images 6. References 1. Lindeberg, T.: Edge Detection and Ridge detection with automatic scale selection. IJCV (1998). 2. Arbelaez, P. et al: Contour detection and hierarchical image segmentation. PAMI (2011). 3. Levinshtein, A. et al: Multiscale symmetric part detection and grouping. ICCV (2009). 1. Outline – Contributions Automatic ridge detection in natural images. Learning-based approach allows flexibility; our detector can be tailored to a specific task. Exploitation of color, brightness, texture and spectral cues. Frontend for higher-level tasks such as object detection and shape representation