Download presentation
Presentation is loading. Please wait.
Published byRudolph Beasley Modified over 9 years ago
1
Multiresolution Histograms and their Use for Texture Classification Stathis Hadjidemetriou, Michael Grossberg and Shree Nayar CAVE Lab, Columbia University Partially funded by NSF ITR Award, DARPA/ONR MURI
2
Fast and Simple Feature Q: Is there a fast feature which captures spatial information? A: Consider multiple resolutions. Same Histogram
3
Histograms of Filtered Images Graylevel Bin Count Graylevel Bin Count Histograms Resolution
4
Analysis of Multiresolution Histograms Bin Count Graylevel Bin Count Change Shape and Texture Properties Difference Histograms Multiresolution Histograms Shape and Texture Images ?
5
Tools for Analysizing the Histogram Shanon Entropy Change in Shanon Entropy: Fisher Information Generalization: Tsallis Entropy/Generalized Fisher Information Resolution Multiresolution Histogram Bin Filter Dependent Constant
6
Relating Histogram Change to Image Fisher Information: Measure of image sharpness [Stam, 59, Plastino et al, 97]: Image Image Gradient Image Domain Edge filter never computed: Implicit
7
Analysis of Multiresolution Histograms Bin Count Graylevel Bin Count Change Shape and Texture Properties Difference Histograms Multiresolution Histograms Shape and Texture Images Fisher Information Resolution Fisher Information Shape Elongation Shape Boundary Texel Repetition Texel Placement
8
Shape Elongation and Fisher Information Elongation: St. dev. along axes: x, y. Gaussian: Sides of base: r x, r y. Pyramid: (analytically) Elongation: 6 5 4 3 2 12345 J
9
Shape Boundary and Fisher Information Superquadrics: =0.56 =1.00 =1.48 =2.00 =6.67 (numerically) Complex boundary J 2 3 4 5 6 2460
10
Texel Repetition and Fisher Information (analytically). 142356 0 2 4 6 8 Tileing 142356 Tileing p J 0 2 4 6 8 J x 10 3
11
Texel Placement and Fisher Information Stand. dev. of perturbation (numerically)Randomness Average of 20 trials 01551020 6.6 6.4 6.2 6 5.8 J St. Dev (% of Texel Width) 01551020 St. Dev (% of Texel Width) x 10 3 2.9 2.8 2.7 2.6 2.5 J
12
L 1 norm Matching Algorithm Multiresolution histogram with Burt-Adelson Pyramid Cumulative histograms Difference histograms between consecutive resolutions Concatenate to form feature vector Compute Feature
13
Histograms Bin Width Histogram bin width: Subsampling factor in pyramid:
14
Parameters of Multiresolution Histogram Histogram smoothing to avoid aliasing: –Database images –Test images Histogram normalization – Image size – Histogram size
15
Databases for Matching Database of CUReT textures [Dana et al, 99]: –8,046 images; 61 materials –Histogram equalized Database of Brodatz textures [Brodatz, 66]: –91 images; 7 images –Histogram equalized
16
Database of Brodatz Textures Samples of equalized images:
17
Match Results for Brodatz Textures Match under Gaussian noise of st.dev. 15 graylevels
18
Class Matching Sensitivity: Brodatz Textures 0102030405060 0 20 40 60 80 100 St dev. of noise n Class matched 8 16 32 62 128 256 Number of bins
19
Class Matching Sensitivity: Brodatz Textures smoothing & adaptive bin size 0102030405060 100 St dev. of noise n 95 90 85 80 75 70 65 60 256 Constant 256, Higher Subsampling= 2 2/3 256, Lower Subsampling = 2 1/2
20
Database of Curet Textures Samples of equalized images:
21
Match Results for Curet Textures Match under Gaussian noise of st.dev. 15 graylevels.
22
Match 100 randomly selected images per noise level Difference norm & Smoothing Class Matching Sensitivity: CUReT Textures 01020304050 60 70 80 90 100 St dev. of noise n Class matched 256 Constant 256, Higher Subsampling= 2 2/3 256, Lower Subsampling = 2 1/2
23
Comparison with Low-level Features Fourier power spectrum annuliFourier power spectrum annuli Gabor featuresGabor features Daubechies wavelet featuresDaubechies wavelet features Auto-cooccurrence matrixAuto-cooccurrence matrix Markov random field parametersMarkov random field parameters
24
Comparison with Low-Level Features Auto-cooccurrence matrix Fourier power spectrum annuli: Gabor features r1r1 r2r2
25
Comparison with Low-Level Features Markov random field parameters Wavelets decomposition Wavelet packets decomposition Wavelet coefficient energies:
26
Comparison of Computation Costs 1 Markov random field parameters O(n( 2 -1) 2 -( 2 -1) 3 /3) 2 Gabor features ( (logn+1)nlogn 1/2 ) 3 Fourier power spectrum features O(n 3/2 ) 4 Auto-cooccurrence matrix O(n ) 5 Wavelet coefficient energies O(n l) 6 Multiresolution histograms n n- number of pixels - window width l- resolution levels decreasing cost
27
Sensitivity Comparison to Transformations FeatureTranslationRotation Uniform Scaling 1 Fourier power spectrum annuli invariantrobustequivariant 2 Gabor features invariantvariantequivariant 3 Daubechies wavelet energies variantvariantvariant 4 Multiresolution histograms invariantinvariantequivariant 5 Auto-cooccurrence matrix invariantrobustequivariant 6 Markov random field parameters invariantvariantvariant
28
Matching Comparison of Features: Brodatz Brodatz textures database: 0102030405060 0 20 40 60 80 100 St dev. of noise n Class matched Multiresolution Diff. Histograms Fourier Power Spectrum Gabor Features Wavelet Packets Cooccurence Matrix Markov Random Fields
29
Matching Comparison of Features: CUReT Curet textures database: Match 100 randomly selected images per noise level 01020304050 St dev. of noise n 0 20 40 60 80 100 Class matched Multiresolution Diff. Histograms Fourier Power Spectrum Gabor Features Wavelet Packets Cooccurence Matrix Markov Random Fields r1r1
30
Sensitivity of Features to Recognition Feature Gaussian Noise Database size,#classes IlluminationParameter selection Fourier power spectrum annuli sensitive robustvery sensitive Gabor featuresrobust sensitive Daubechies wavelet energies sensitiverobust Multiresolution histogram robust Auto-cooccurrence matrix very sensitive Markov random field parameters very sensitive sensitiveN/A
31
Colors
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.