Evaluation Of Color Descriptors For Object And Scene
Download as PPTX, PDF2 likes1,497 views
This document evaluates different color descriptors for object and scene recognition. It finds that descriptors like W-SIFT and rgSIFT, which are invariant to light intensity scale and shifts, generally outperform other descriptors. However, the usefulness of invariance is found to be category-specific, as some objects like cars and tables do not benefit from light intensity invariance. Evaluation on image and video benchmarks shows scale and shift invariant descriptors work best for most but not all categories.
Evaluation Of Color Descriptors For Object And Scene
- 1. Evaluation of Color Descriptors for Object and Scene RecognitionAuthors: Koen E. A. van de Sande, Theo Gevers, and Cees G. M. Snoek @ University of AmsterdamPresenter: Shao-Chuan Wang
- 2. Evaluation of Color Descriptors for Object and Scene RecognitionFocus: Color features/descriptors on obj. and scene recognitionSummary: The invariance of photometric transform aces and its effect on discriminative power.Conclusion: The usefulness of invariance is category-specific!
- 3. Photometric transforms (1/2)Light intensity scale invariantLight intensity shift invariant
- 4. Photometric transforms (1/2)Light intensity scale and shift invariantLight color changeLight color change and shift
- 5. Color Descriptors (1/1)Histograms RGB, Hue, Saturation, rgHistogram, …Color Moments: contain spatial info.Color SIFT: combined color and SIFTHSV-SIFT, Hue-SIFT, …
- 6. Color Histograms (1/2)RGB-histogramHue-histogramH and S are scale-invariant and shift-invariant w.r.t light intensityrg-histogramScale-invariantNot shift-invariantNot that b is redudantImage from wikipedia
- 7. Color Histograms (2/2)Transformed colorScale and shift-invariant w.r.t light intensity.Opponent color histogramO1,O2 shift invariantO3: intensity, no invariant
- 8. Color SIFT Descriptors (1/3)HSV-SIFTSIFT over HSV channelsHue is unstable in gray axisHue-SIFT (Van de Weijer 2006)Used Hue histogram weighing by its saturationConcatenate Hue histogram with SIFTOnly SIFT is invariant; Hue histogram is not! (partial invariance)Hue Instability
- 9. Color SIFT Descriptors (2/3)OpponentSIFTSIFT over all channels in the opponent color space.Shift-invariant to light intensity.W-SIFTEliminate O1 and O2’s intensity informationScale-invariant to light intensityrg-SIFTSIFT over r,g spacesScale and shift invariant, but not invariant to light color changes/shifts
- 10. Color SIFT Descriptors (3/3)Transformed color SIFTSIFT over normalized transformed channels.Scale- and shift-invariant to light color changes and shift.
- 11. ExperimentsImplementation:Scale-invariants points are detected by Harris-Laplace point detectorsColor descriptors are computed over the area around the points; all regions are proportionally re-sampled to a uniform 60x60 patch.Cluster descriptors with k = 40 (images) k = 4000 (video)SVM classifier with EMD/chi-square kernel
- 12. Benchmark (1/3)Image: PASCAL VOC 2007, over 20 object categories
- 13. Benchmark (2/3)Most objs were categorized better under scale- and shift- invariant to light intensitySome, such as car and dining table, do not benefit from such invariance.
- 14. Benchmark (3/3)Video: Mediamill Challenge, 39 object and scene categories
- 15. Evaluation of Color Descriptors for Object and Scene RecognitionConclusion:W-SIFT and rgSIFT, in general, outperform other color descriptors.Light intensity info. Is important for some categoriesUsefulness of invariance is category-specific.