Contourlet Transforms For Feature DetectionFeature DetectionContourlets (Do and Vetterli, 2005)Contourlet filter bankTest Pattern Image – Scale 1Test Pattern Image – Scale 2Peppers Image – Scale 1Peppers Image – Scale 2Generic Girl Image – Scale 1Generic Girl Image – Scale 2Tiffany Image – Scale 1Tiffany Image – Scale 2Elaine Image – Scale 1Elaine Image – Scale 2Lena Image – Scale 1Lena Image – Scale 2ConclusionsContourlet Transforms For Feature DetectionWei-shi TsaiApril 29th, 2008Feature DetectionFocus will be on edge detectionGradient operators (Sobel, Roberts)Laplacian operatorsLoG (Laplacian of Gaussian)DoG (Difference of Gaussians)Canny methodAnisotropic diffusionContourlets (Do and Vetterli, 2005)Captures smooth contours and edges at any orientationFilters noiseDerived directly from discrete domain instead of extending from continuous domainCan be implemented using filter banksContourlet filter bankThe transform decouples the multiscale and the directional decompositions.Test Pattern Image – Scale 1Test Pattern Image – Scale 2Peppers Image – Scale 1Peppers Image – Scale 2Generic Girl Image – Scale 1Generic Girl Image – Scale 2Tiffany Image – Scale 1Tiffany Image – Scale 2Elaine Image – Scale 1Elaine Image – Scale 2Lena Image – Scale 1Lena Image – Scale 2ConclusionsContourlet transforms can be used for edge detectionResults can vary based on the type of imageEvaluation is only useful given what the feature extracted is to be
View Full Document